initial commit

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2022 Omri Avrahami
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

configs/__init__.py

@@ -0,0 +1,90 @@
+import argparse
+import os
+
+import yaml
+
+__all__ = ['get_config', 'print_config']
+
+
+def get_config(args):
+
+    config = dict2namespace(setdefault(_get_raw_config(args.config), _get_raw_config("default.yml")))
+
+    if not hasattr(config.sampling, "sigma_dist"):
+        config.sampling.sigma_dist = config.model.sigma_dist
+    if not hasattr(config.biggan, "resolution"):
+        config.biggan.resolution = config.data.image_size
+
+    if args.consistent:
+        config.sampling.consistent = args.consistent
+        config.sampling.noise_first = False
+    if args.step_lr:
+        config.sampling.step_lr = args.step_lr
+    if args.nsigma != 0:
+        config.sampling.nsigma = args.nsigma
+    if args.step_lr != 0:
+        config.sampling.step_lr = args.step_lr
+    if args.batch_size != 0:
+        config.sampling.batch_size = args.batch_size
+        config.fast_fid.batch_size = args.batch_size
+    # ToDo: experimental and is only using model_types
+    if args.model_types is not None and len(args.model_types)==1 and args.model_types[0] in [0, 6, 23] and config.data.dataset in ['tinyImages', 'CIFAR10']:
+        config.sampling.batch_size = min(200, config.sampling.batch_size)
+    if args.model_types is not None and len(args.model_types) == 1 and args.model_types[0] in [8] and config.data.dataset in ['tinyImages', 'CIFAR10']:
+        config.sampling.batch_size = min(800, config.sampling.batch_size)
+
+    if args.ODI_steps == -1:
+        args.ODI_steps = None
+    if args.fid_num_samples != 0:
+        config.fast_fid.num_samples = args.fid_num_samples
+    if args.begin_ckpt != 0:
+        config.fast_fid.begin_ckpt = args.begin_ckpt
+        config.sampling.ckpt_id = args.begin_ckpt
+    if args.end_ckpt != 0:
+        config.fast_fid.end_ckpt = args.begin_ckpt
+    if args.adam:
+        config.optim.beta1 = args.adam_beta[0]
+        config.optim.beta2 = args.adam_beta[1]
+    if args.D_adam:
+        config.optim.adv_beta1 = args.D_adam_beta[0]
+        config.optim.adv_beta2 = args.D_adam_beta[1]
+    if args.D_steps != 0:
+        config.adversarial.D_steps = args.D_steps
+
+    return config
+
+
+def _get_raw_config(name):
+    here = os.path.dirname(os.path.abspath(__file__))
+    with open(os.path.join(here, name), 'r') as f:
+        yaml_dict = yaml.load(f, Loader=yaml.FullLoader)
+    return yaml_dict
+
+
+def setdefault(config, default):
+    #print('config is', config, 'default is', default)
+    for x in default:
+        v = default.get(x)
+        if isinstance(v, dict) and x in config:
+            setdefault(config.get(x), v)
+        else:
+            config.setdefault(x, v)
+    return config
+
+
+def dict2namespace(config):
+    namespace = argparse.Namespace()
+    for key, value in config.items():
+        if isinstance(value, dict):
+            new_value = dict2namespace(value)
+        else:
+            new_value = value
+        setattr(namespace, key, new_value)
+    return namespace
+
+
+def print_config(config):
+    print(">" * 80)
+    print(yaml.dump(config, default_flow_style=False))
+    print("<" * 80)
+

configs/default.yml

@@ -0,0 +1,70 @@
+data:
+  dataset: LSUN
+  category: church_outdoor
+  image_size: 64
+  channels: 3
+  logit_transform: false
+  uniform_dequantization: false
+  gaussian_dequantization: false
+  random_flip: true
+  rescaled: false
+  num_workers: 32
+
+model:
+  sigma_begin: 140
+  num_classes: 788
+  ema: true
+  ema_rate: 0.999
+  spec_norm: false
+  sigma_dist: geometric
+  sigma_end: 0.01
+  normalization: InstanceNorm++
+  nonlinearity: elu
+  ngf: 128
+  unet: false
+  dropout: 0.0
+
+sampling:
+  noise_first: false
+  clamp: false
+  experiment_name: 'logits_evolution_tracking_best_point'
+
+training:
+  log_all_sigmas: true  # whether to send training information to tensorboard
+
+optim:
+  weight_decay: 0.000
+  optimizer: Adam
+  lr: 0.0001
+  beta1: 0.9
+  beta2: 0.999
+  adv_beta1: -0.5
+  adv_beta2: 0.9
+  amsgrad: false
+  eps: 0.00000001
+  momentum: 0.9
+
+fast_fid:
+  batch_size: 50
+  num_samples: 1000
+
+adversarial:
+  lambda_dae: 1           # multiplier term for Lp loss function (only used in GAN setting)
+  lambda_D: 1             # multiplier term for GAN loss of the discriminator'
+  lambda_G_gan: 1         # multiplier term for GAN loss of the generator
+  D_steps: 1              # Discriminator steps per Generator step
+  adv_loss: LSGAN         # 'GAN, LSGAN, HingeGAN, RpGAN, RaGAN, RaLSGAN, RaHingeGAN'
+  arch: 2                 # 0 is DCGAN_D0, 1 is DCGAN_D1, 2 is BigGAN
+  spectral: false         # If True, spectral normalization in D
+  no_batch_norm_D: false  # Not active in BigGAN
+  adv_clamp: true     # If True, do not clamp output of score network before giving to discriminator
+
+biggan:
+  ch: 64          # Number of channels
+  thin: false     # If True, use thin Discriminator (D_ch = 0 with D_thin = True leads to )
+  kernel_size: 3
+  attn: '64'       # Number of attention filters (If 0, do not use self-attention)
+  n_classes: 1    # Number of classes of the dataset (If = 1 leads to Unconditional GAN)  # inactive
+  init: xavier    # Type of init,: ortho, xavier, N02
+
+

configs/imagenet1000.yml

@@ -0,0 +1,196 @@
+training:
+  batch_size: 128
+  n_epochs: 500000
+  n_iters: 300001
+  snapshot_freq: 5000
+  snapshot_sampling: false
+  snapshot_sampling_freq: 5000
+  anneal_power: 2
+  log_all_sigmas: false
+
+sampling:
+  batch_size: 100 # 25 for randomized smoothing, 100 otherwise
+  data_init: false # true for CLIP
+  data_init_correct_class: false
+  random_label_data_init: false # true for CLIP
+  model_failure_examples_init: true # false for CLIP
+  model_failure_examples_init_wrong_class: false
+  name_init: false
+  init_data_seed: 1234 # set it to the same seed (1234)! 1239 for CLIP
+  step_lr: 0.0000062
+  nsigma: 5
+  ckpt_id: 260000
+  final_only: false
+  fid: false
+  num_samples4fid: 10000
+  inpainting: false
+  interpolation: false
+  n_interpolations: 15
+  noise_first: true
+  save_freq: 1
+  consistent: false
+  conditional: true
+  model_description:
+    type: ['benchmark-Madry_l2_experimental'] # benchmark-Madry_l2_improved, benchmark-Madry_linf_improved,
+                                         # benchmark-Madry_l2_experimental, benchmark-Madry_linf_experimental
+                                         # benchmark-Madry_l2_improved_eps_1
+
+    folder: 'AdvACET_24-02-2020_14:41:39'
+    checkpoint: 'final'
+    temperature: null # 0.41
+    load_temp: false
+  line_search:
+    type: 'armijo' # 'wolfe_conditions'  # 'armijo' # 'wolfe_conditions' # wolfe_conditions, armijo_momentum_prox, armijo
+    beta_1: 0.0001
+    beta_2: 0.9999
+    alpha_lower_bound: 0.0001
+    alpha_current: 100.0 # 1000, change it to 100
+  inverse_mode:
+    noise: 'gaussian'
+    init_apgd_eps: 3.9
+    init_apgd_steps: 5
+    n_restarts: 5
+    activate: false
+    type: 'not_dynamic_penalty' # alma, not_dynamic_penalty
+    multiplier: 10000 # 1000, 10 not used in current, augmented Lagrangian setting - ToDo - delete?
+    inverse_mode_threshold_probs: 0.99974
+    norm: 'l2' # l2, l1, LPIPS, 1.5, FeatureDist
+    #  further hyper-parameters according to https://github.com/jeromerony/adversarial-library/blob/main/adv_lib/attacks/augmented_lagrangian_attacks/alma.py
+    penalty_multiplier_init: 1
+    lagrange_multiplier_init: 1
+    lr_init: 0.1 # 0.1
+    is_adv: null
+    penalty_multiplier: null
+    lagrange_multiplier: null
+    lr_reduction: 0.1
+    constr_improvement_rate: 0.95
+    prev_constraint: null
+    ema_weight: 0.999
+    penalty_param_increase: 10 #1.2
+    check_steps: 10
+    init_lr_distance: null # 0.1
+    # RMSProp-related constants
+    square_avg: null
+    # ToDo: what is a better initial value for RMSProp? 0 vs 1?
+    square_avg_init: 0 # 1
+    alpha: 0.99
+    eps: 1e-9 # 1e-8
+    # Adam-related constants
+    beta_1: 0.9
+    beta_2: 0.999
+  ratio_mode:
+    no_apgd: false
+    frank_wolfe:
+      decay: false
+      activate: false
+      constraint: 'intersection'
+      momentum: 0 # 0.9
+      backtracking_LS:
+        activate: false
+        L: null
+        eta: 0.9 # 0.9
+        tau: 2 # 2
+        gamma_max: 1
+        eps: 0.001
+    randomized_smoothing: False
+    randomized_smoothing_sigma: 0.25 # 0.25
+    activate: false
+    noise: false # gaussian, uniform, false
+    loss_diverse_iter: false # false, 20
+    eps_repeat_steps: 2 # 8
+    step_lr: 0.5 # 0.1
+    adaptive_stepsize: false
+    momentum: false
+    grad_normalization: 'l2' # false, l1, l2, l_inf
+    apgd:
+      pgd_mode: false
+      activate: true
+      n_restarts: 5
+      p_current: 0
+      p_next: 0.22
+      lr_current: 0.5 # 0.5, 0.9, 10
+      lr_next: 0.5 # 0.5, 0.9, 10
+      lr_next_for_plotting: 0.5 # 0.5, 0.9, 10
+      period_length_red: 0.03
+      min_length: 0.06
+      loss_values: [ ]
+      f_max_current: 0
+      f_max_next: 0
+      rho: 0.75
+      alpha: 0.75
+  use_generative_model: false
+  grad_scale: 10000
+  logits_max: false
+  use_noise: false
+  project: true
+  eps_projection: 1 # for generating misalignment for score networks, when misalignment == true
+  eps_begin: 12 # 25 for CLIP
+  eps_end: 12 # 25 for CLIP
+  experiment_name: 'logits_evolution' #'logits_evolution_FID_imagenet1000'
+  save_final_name: false # 'final-most_probable_ood_init_RATIO_232steps' #
+  start_with_final_name: false # 'final-most_probable_ood_init_RATIO_232steps', 'final_ood_init_RATIO_232steps'
+  norm_projection: 'l2' # 'LPIPS', 'l1', 'l2'
+  regularizer: false # false, 'LPIPS', 'l1', 'l2'
+  misalignment: false
+  calculade_FID_stats: false # true except for FID calculation set to true
+  calculade_FID_continue: false
+  calculate_FID_continue_path: ''
+
+fast_fid:
+  batch_size: 1000
+  num_samples: 1000
+  begin_ckpt: 5000
+  end_ckpt: 300000
+
+data:
+  dataset: "ImageNet1000" # oa-imagenet, CIFAR10
+  dataset_for_scorenet: 'ImageNet1000'
+  class_labels: ['tench, Tinca tinca', 'goldfish, Carassius auratus', 'great white shark, white shark, man-eater, man-eating shark, Carcharodon carcharias', 'tiger shark, Galeocerdo cuvieri', 'hammerhead, hammerhead shark', 'electric ray, crampfish, numbfish, torpedo', 'stingray', 'cock', 'hen', 'ostrich, Struthio camelus', 'brambling, Fringilla montifringilla', 'goldfinch, Carduelis carduelis', 'house finch, linnet, Carpodacus mexicanus', 'junco, snowbird', 'indigo bunting, indigo finch, indigo bird, Passerina cyanea', 'robin, American robin, Turdus migratorius', 'bulbul', 'jay', 'magpie', 'chickadee', 'water ouzel, dipper', 'kite', 'bald eagle, American eagle, Haliaeetus leucocephalus', 'vulture', 'great grey owl, great gray owl, Strix nebulosa', 'European fire salamander, Salamandra salamandra', 'common newt, Triturus vulgaris', 'eft', 'spotted salamander, Ambystoma maculatum', 'axolotl, mud puppy, Ambystoma mexicanum', 'bullfrog, Rana catesbeiana', 'tree frog, tree-frog', 'tailed frog, bell toad, ribbed toad, tailed toad, Ascaphus trui', 'loggerhead, loggerhead turtle, Caretta caretta', 'leatherback turtle, leatherback, leathery turtle, Dermochelys coriacea', 'mud turtle', 'terrapin', 'box turtle, box tortoise', 'banded gecko', 'common iguana, iguana, Iguana iguana', 'American chameleon, anole, Anolis carolinensis', 'whiptail, whiptail lizard', 'agama', 'frilled lizard, Chlamydosaurus kingi', 'alligator lizard', 'Gila monster, Heloderma suspectum', 'green lizard, Lacerta viridis', 'African chameleon, Chamaeleo chamaeleon', 'Komodo dragon, Komodo lizard, dragon lizard, giant lizard, Varanus komodoensis', 'African crocodile, Nile crocodile, Crocodylus niloticus', 'American alligator, Alligator mississipiensis', 'triceratops', 'thunder snake, worm snake, Carphophis amoenus', 'ringneck snake, ring-necked snake, ring snake', 'hognose snake, puff adder, sand viper', 'green snake, grass snake', 'king snake, kingsnake', 'garter snake, grass snake', 'water snake', 'vine snake', 'night snake, Hypsiglena torquata', 'boa constrictor, Constrictor constrictor', 'rock python, rock snake, Python sebae', 'Indian cobra, Naja naja', 'green mamba', 'sea snake', 'horned viper, cerastes, sand viper, horned asp, Cerastes cornutus', 'diamondback, diamondback rattlesnake, Crotalus adamanteus', 'sidewinder, horned rattlesnake, Crotalus cerastes', 'trilobite', 'harvestman, daddy longlegs, Phalangium opilio', 'scorpion', 'black and gold garden spider, Argiope aurantia', 'barn spider, Araneus cavaticus', 'garden spider, Aranea diademata', 'black widow, Latrodectus mactans', 'tarantula', 'wolf spider, hunting spider', 'tick', 'centipede', 'black grouse', 'ptarmigan', 'ruffed grouse, partridge, Bonasa umbellus', 'prairie chicken, prairie grouse, prairie fowl', 'peacock', 'quail', 'partridge', 'African grey, African gray, Psittacus erithacus', 'macaw', 'sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita', 'lorikeet', 'coucal', 'bee eater', 'hornbill', 'hummingbird', 'jacamar', 'toucan', 'drake', 'red-breasted merganser, Mergus serrator', 'goose', 'black swan, Cygnus atratus', 'tusker', 'echidna, spiny anteater, anteater', 'platypus, duckbill, duckbilled platypus, duck-billed platypus, Ornithorhynchus anatinus', 'wallaby, brush kangaroo', 'koala, koala bear, kangaroo bear, native bear, Phascolarctos cinereus', 'wombat', 'jellyfish', 'sea anemone, anemone', 'brain coral', 'flatworm, platyhelminth', 'nematode, nematode worm, roundworm', 'conch', 'snail', 'slug', 'sea slug, nudibranch', 'chiton, coat-of-mail shell, sea cradle, polyplacophore', 'chambered nautilus, pearly nautilus, nautilus', 'Dungeness crab, Cancer magister', 'rock crab, Cancer irroratus', 'fiddler crab', 'king crab, Alaska crab, Alaskan king crab, Alaska king crab, Paralithodes camtschatica', 'American lobster, Northern lobster, Maine lobster, Homarus americanus', 'spiny lobster, langouste, rock lobster, crawfish, crayfish, sea crawfish', 'crayfish, crawfish, crawdad, crawdaddy', 'hermit crab', 'isopod', 'white stork, Ciconia ciconia', 'black stork, Ciconia nigra', 'spoonbill', 'flamingo', 'little blue heron, Egretta caerulea', 'American egret, great white heron, Egretta albus', 'bittern', 'crane', 'limpkin, Aramus pictus', 'European gallinule, Porphyrio porphyrio', 'American coot, marsh hen, mud hen, water hen, Fulica americana', 'bustard', 'ruddy turnstone, Arenaria interpres', 'red-backed sandpiper, dunlin, Erolia alpina', 'redshank, Tringa totanus', 'dowitcher', 'oystercatcher, oyster catcher', 'pelican', 'king penguin, Aptenodytes patagonica', 'albatross, mollymawk', 'grey whale, gray whale, devilfish, Eschrichtius gibbosus, Eschrichtius robustus', 'killer whale, killer, orca, grampus, sea wolf, Orcinus orca', 'dugong, Dugong dugon', 'sea lion', 'Chihuahua', 'Japanese spaniel', 'Maltese dog, Maltese terrier, Maltese', 'Pekinese, Pekingese, Peke', 'Shih-Tzu', 'Blenheim spaniel', 'papillon', 'toy terrier', 'Rhodesian ridgeback', 'Afghan hound, Afghan', 'basset, basset hound', 'beagle', 'bloodhound, sleuthhound', 'bluetick', 'black-and-tan coonhound', 'Walker hound, Walker foxhound', 'English foxhound', 'redbone', 'borzoi, Russian wolfhound', 'Irish wolfhound', 'Italian greyhound', 'whippet', 'Ibizan hound, Ibizan Podenco', 'Norwegian elkhound, elkhound', 'otterhound, otter hound', 'Saluki, gazelle hound', 'Scottish deerhound, deerhound', 'Weimaraner', 'Staffordshire bullterrier, Staffordshire bull terrier', 'American Staffordshire terrier, Staffordshire terrier, American pit bull terrier, pit bull terrier', 'Bedlington terrier', 'Border terrier', 'Kerry blue terrier', 'Irish terrier', 'Norfolk terrier', 'Norwich terrier', 'Yorkshire terrier', 'wire-haired fox terrier', 'Lakeland terrier', 'Sealyham terrier, Sealyham', 'Airedale, Airedale terrier', 'cairn, cairn terrier', 'Australian terrier', 'Dandie Dinmont, Dandie Dinmont terrier', 'Boston bull, Boston terrier', 'miniature schnauzer', 'giant schnauzer', 'standard schnauzer', 'Scotch terrier, Scottish terrier, Scottie', 'Tibetan terrier, chrysanthemum dog', 'silky terrier, Sydney silky', 'soft-coated wheaten terrier', 'West Highland white terrier', 'Lhasa, Lhasa apso', 'flat-coated retriever', 'curly-coated retriever', 'golden retriever', 'Labrador retriever', 'Chesapeake Bay retriever', 'German short-haired pointer', 'vizsla, Hungarian pointer', 'English setter', 'Irish setter, red setter', 'Gordon setter', 'Brittany spaniel', 'clumber, clumber spaniel', 'English springer, English springer spaniel', 'Welsh springer spaniel', 'cocker spaniel, English cocker spaniel, cocker', 'Sussex spaniel', 'Irish water spaniel', 'kuvasz', 'schipperke', 'groenendael', 'malinois', 'briard', 'kelpie', 'komondor', 'Old English sheepdog, bobtail', 'Shetland sheepdog, Shetland sheep dog, Shetland', 'collie', 'Border collie', 'Bouvier des Flandres, Bouviers des Flandres', 'Rottweiler', 'German shepherd, German shepherd dog, German police dog, alsatian', 'Doberman, Doberman pinscher', 'miniature pinscher', 'Greater Swiss Mountain dog', 'Bernese mountain dog', 'Appenzeller', 'EntleBucher', 'boxer', 'bull mastiff', 'Tibetan mastiff', 'French bulldog', 'Great Dane', 'Saint Bernard, St Bernard', 'Eskimo dog, husky', 'malamute, malemute, Alaskan malamute', 'Siberian husky', 'dalmatian, coach dog, carriage dog', 'affenpinscher, monkey pinscher, monkey dog', 'basenji', 'pug, pug-dog', 'Leonberg', 'Newfoundland, Newfoundland dog', 'Great Pyrenees', 'Samoyed, Samoyede', 'Pomeranian', 'chow, chow chow', 'keeshond', 'Brabancon griffon', 'Pembroke, Pembroke Welsh corgi', 'Cardigan, Cardigan Welsh corgi', 'toy poodle', 'miniature poodle', 'standard poodle', 'Mexican hairless', 'timber wolf, grey wolf, gray wolf, Canis lupus', 'white wolf, Arctic wolf, Canis lupus tundrarum', 'red wolf, maned wolf, Canis rufus, Canis niger', 'coyote, prairie wolf, brush wolf, Canis latrans', 'dingo, warrigal, warragal, Canis dingo', 'dhole, Cuon alpinus', 'African hunting dog, hyena dog, Cape hunting dog, Lycaon pictus', 'hyena, hyaena', 'red fox, Vulpes vulpes', 'kit fox, Vulpes macrotis', 'Arctic fox, white fox, Alopex lagopus', 'grey fox, gray fox, Urocyon cinereoargenteus', 'tabby, tabby cat', 'tiger cat', 'Persian cat', 'Siamese cat, Siamese', 'Egyptian cat', 'cougar, puma, catamount, mountain lion, painter, panther, Felis concolor', 'lynx, catamount', 'leopard, Panthera pardus', 'snow leopard, ounce, Panthera uncia', 'jaguar, panther, Panthera onca, Felis onca', 'lion, king of beasts, Panthera leo', 'tiger, Panthera tigris', 'cheetah, chetah, Acinonyx jubatus', 'brown bear, bruin, Ursus arctos', 'American black bear, black bear, Ursus americanus, Euarctos americanus', 'ice bear, polar bear, Ursus Maritimus, Thalarctos maritimus', 'sloth bear, Melursus ursinus, Ursus ursinus', 'mongoose', 'meerkat, mierkat', 'tiger beetle', 'ladybug, ladybeetle, lady beetle, ladybird, ladybird beetle', 'ground beetle, carabid beetle', 'long-horned beetle, longicorn, longicorn beetle', 'leaf beetle, chrysomelid', 'dung beetle', 'rhinoceros beetle', 'weevil', 'fly', 'bee', 'ant, emmet, pismire', 'grasshopper, hopper', 'cricket', 'walking stick, walkingstick, stick insect', 'cockroach, roach', 'mantis, mantid', 'cicada, cicala', 'leafhopper', 'lacewing, lacewing fly', "dragonfly, darning needle, devil's darning needle, sewing needle, snake feeder, snake doctor, mosquito hawk, skeeter hawk", 'damselfly', 'admiral', 'ringlet, ringlet butterfly', 'monarch, monarch butterfly, milkweed butterfly, Danaus plexippus', 'cabbage butterfly', 'sulphur butterfly, sulfur butterfly', 'lycaenid, lycaenid butterfly', 'starfish, sea star', 'sea urchin', 'sea cucumber, holothurian', 'wood rabbit, cottontail, cottontail rabbit', 'hare', 'Angora, Angora rabbit', 'hamster', 'porcupine, hedgehog', 'fox squirrel, eastern fox squirrel, Sciurus niger', 'marmot', 'beaver', 'guinea pig, Cavia cobaya', 'sorrel', 'zebra', 'hog, pig, grunter, squealer, Sus scrofa', 'wild boar, boar, Sus scrofa', 'warthog', 'hippopotamus, hippo, river horse, Hippopotamus amphibius', 'ox', 'water buffalo, water ox, Asiatic buffalo, Bubalus bubalis', 'bison', 'ram, tup', 'bighorn, bighorn sheep, cimarron, Rocky Mountain bighorn, Rocky Mountain sheep, Ovis canadensis', 'ibex, Capra ibex', 'hartebeest', 'impala, Aepyceros melampus', 'gazelle', 'Arabian camel, dromedary, Camelus dromedarius', 'llama', 'weasel', 'mink', 'polecat, fitch, foulmart, foumart, Mustela putorius', 'black-footed ferret, ferret, Mustela nigripes', 'otter', 'skunk, polecat, wood pussy', 'badger', 'armadillo', 'three-toed sloth, ai, Bradypus tridactylus', 'orangutan, orang, orangutang, Pongo pygmaeus', 'gorilla, Gorilla gorilla', 'chimpanzee, chimp, Pan troglodytes', 'gibbon, Hylobates lar', 'siamang, Hylobates syndactylus, Symphalangus syndactylus', 'guenon, guenon monkey', 'patas, hussar monkey, Erythrocebus patas', 'baboon', 'macaque', 'langur', 'colobus, colobus monkey', 'proboscis monkey, Nasalis larvatus', 'marmoset', 'capuchin, ringtail, Cebus capucinus', 'howler monkey, howler', 'titi, titi monkey', 'spider monkey, Ateles geoffroyi', 'squirrel monkey, Saimiri sciureus', 'Madagascar cat, ring-tailed lemur, Lemur catta', 'indri, indris, Indri indri, Indri brevicaudatus', 'Indian elephant, Elephas maximus', 'African elephant, Loxodonta africana', 'lesser panda, red panda, panda, bear cat, cat bear, Ailurus fulgens', 'giant panda, panda, panda bear, coon bear, Ailuropoda melanoleuca', 'barracouta, snoek', 'eel', 'coho, cohoe, coho salmon, blue jack, silver salmon, Oncorhynchus kisutch', 'rock beauty, Holocanthus tricolor', 'anemone fish', 'sturgeon', 'gar, garfish, garpike, billfish, Lepisosteus osseus', 'lionfish', 'puffer, pufferfish, blowfish, globefish', 'abacus', 'abaya', "academic gown, academic robe, judge's robe", 'accordion, piano accordion, squeeze box', 'acoustic guitar', 'aircraft carrier, carrier, flattop, attack aircraft carrier', 'airliner', 'airship, dirigible', 'altar', 'ambulance', 'amphibian, amphibious vehicle', 'analog clock', 'apiary, bee house', 'apron', 'ashcan, trash can, garbage can, wastebin, ash bin, ash-bin, ashbin, dustbin, trash barrel, trash bin', 'assault rifle, assault gun', 'backpack, back pack, knapsack, packsack, rucksack, haversack', 'bakery, bakeshop, bakehouse', 'balance beam, beam', 'balloon', 'ballpoint, ballpoint pen, ballpen, Biro', 'Band Aid', 'banjo', 'bannister, banister, balustrade, balusters, handrail', 'barbell', 'barber chair', 'barbershop', 'barn', 'barometer', 'barrel, cask', 'barrow, garden cart, lawn cart, wheelbarrow', 'baseball', 'basketball', 'bassinet', 'bassoon', 'bathing cap, swimming cap', 'bath towel', 'bathtub, bathing tub, bath, tub', 'beach wagon, station wagon, wagon, estate car, beach waggon, station waggon, waggon', 'beacon, lighthouse, beacon light, pharos', 'beaker', 'bearskin, busby, shako', 'beer bottle', 'beer glass', 'bell cote, bell cot', 'bib', 'bicycle-built-for-two, tandem bicycle, tandem', 'bikini, two-piece', 'binder, ring-binder', 'binoculars, field glasses, opera glasses', 'birdhouse', 'boathouse', 'bobsled, bobsleigh, bob', 'bolo tie, bolo, bola tie, bola', 'bonnet, poke bonnet', 'bookcase', 'bookshop, bookstore, bookstall', 'bottlecap', 'bow', 'bow tie, bow-tie, bowtie', 'brass, memorial tablet, plaque', 'brassiere, bra, bandeau', 'breakwater, groin, groyne, mole, bulwark, seawall, jetty', 'breastplate, aegis, egis', 'broom', 'bucket, pail', 'buckle', 'bulletproof vest', 'bullet train, bullet', 'butcher shop, meat market', 'cab, hack, taxi, taxicab', 'caldron, cauldron', 'candle, taper, wax light', 'cannon', 'canoe', 'can opener, tin opener', 'cardigan', 'car mirror', 'carousel, carrousel, merry-go-round, roundabout, whirligig', "carpenter's kit, tool kit", 'carton', 'car wheel', 'cash machine, cash dispenser, automated teller machine, automatic teller machine, automated teller, automatic teller, ATM', 'cassette', 'cassette player', 'castle', 'catamaran', 'CD player', 'cello, violoncello', 'cellular telephone, cellular phone, cellphone, cell, mobile phone', 'chain', 'chainlink fence', 'chain mail, ring mail, mail, chain armor, chain armour, ring armor, ring armour', 'chain saw, chainsaw', 'chest', 'chiffonier, commode', 'chime, bell, gong', 'china cabinet, china closet', 'Christmas stocking', 'church, church building', 'cinema, movie theater, movie theatre, movie house, picture palace', 'cleaver, meat cleaver, chopper', 'cliff dwelling', 'cloak', 'clog, geta, patten, sabot', 'cocktail shaker', 'coffee mug', 'coffeepot', 'coil, spiral, volute, whorl, helix', 'combination lock', 'computer keyboard, keypad', 'confectionery, confectionary, candy store', 'container ship, containership, container vessel', 'convertible', 'corkscrew, bottle screw', 'cornet, horn, trumpet, trump', 'cowboy boot', 'cowboy hat, ten-gallon hat', 'cradle', 'crane', 'crash helmet', 'crate', 'crib, cot', 'Crock Pot', 'croquet ball', 'crutch', 'cuirass', 'dam, dike, dyke', 'desk', 'desktop computer', 'dial telephone, dial phone', 'diaper, nappy, napkin', 'digital clock', 'digital watch', 'dining table, board', 'dishrag, dishcloth', 'dishwasher, dish washer, dishwashing machine', 'disk brake, disc brake', 'dock, dockage, docking facility', 'dogsled, dog sled, dog sleigh', 'dome', 'doormat, welcome mat', 'drilling platform, offshore rig', 'drum, membranophone, tympan', 'drumstick', 'dumbbell', 'Dutch oven', 'electric fan, blower', 'electric guitar', 'electric locomotive', 'entertainment center', 'envelope', 'espresso maker', 'face powder', 'feather boa, boa', 'file, file cabinet, filing cabinet', 'fireboat', 'fire engine, fire truck', 'fire screen, fireguard', 'flagpole, flagstaff', 'flute, transverse flute', 'folding chair', 'football helmet', 'forklift', 'fountain', 'fountain pen', 'four-poster', 'freight car', 'French horn, horn', 'frying pan, frypan, skillet', 'fur coat', 'garbage truck, dustcart', 'gasmask, respirator, gas helmet', 'gas pump, gasoline pump, petrol pump, island dispenser', 'goblet', 'go-kart', 'golf ball', 'golfcart, golf cart', 'gondola', 'gong, tam-tam', 'gown', 'grand piano, grand', 'greenhouse, nursery, glasshouse', 'grille, radiator grille', 'grocery store, grocery, food market, market', 'guillotine', 'hair slide', 'hair spray', 'half track', 'hammer', 'hamper', 'hand blower, blow dryer, blow drier, hair dryer, hair drier', 'hand-held computer, hand-held microcomputer', 'handkerchief, hankie, hanky, hankey', 'hard disc, hard disk, fixed disk', 'harmonica, mouth organ, harp, mouth harp', 'harp', 'harvester, reaper', 'hatchet', 'holster', 'home theater, home theatre', 'honeycomb', 'hook, claw', 'hoopskirt, crinoline', 'horizontal bar, high bar', 'horse cart, horse-cart', 'hourglass', 'iPod', 'iron, smoothing iron', "jack-o'-lantern", 'jean, blue jean, denim', 'jeep, landrover', 'jersey, T-shirt, tee shirt', 'jigsaw puzzle', 'jinrikisha, ricksha, rickshaw', 'joystick', 'kimono', 'knee pad', 'knot', 'lab coat, laboratory coat', 'ladle', 'lampshade, lamp shade', 'laptop, laptop computer', 'lawn mower, mower', 'lens cap, lens cover', 'letter opener, paper knife, paperknife', 'library', 'lifeboat', 'lighter, light, igniter, ignitor', 'limousine, limo', 'liner, ocean liner', 'lipstick, lip rouge', 'Loafer', 'lotion', 'loudspeaker, speaker, speaker unit, loudspeaker system, speaker system', "loupe, jeweler's loupe", 'lumbermill, sawmill', 'magnetic compass', 'mailbag, postbag', 'mailbox, letter box', 'maillot', 'maillot, tank suit', 'manhole cover', 'maraca', 'marimba, xylophone', 'mask', 'matchstick', 'maypole', 'maze, labyrinth', 'measuring cup', 'medicine chest, medicine cabinet', 'megalith, megalithic structure', 'microphone, mike', 'microwave, microwave oven', 'military uniform', 'milk can', 'minibus', 'miniskirt, mini', 'minivan', 'missile', 'mitten', 'mixing bowl', 'mobile home, manufactured home', 'Model T', 'modem', 'monastery', 'monitor', 'moped', 'mortar', 'mortarboard', 'mosque', 'mosquito net', 'motor scooter, scooter', 'mountain bike, all-terrain bike, off-roader', 'mountain tent', 'mouse, computer mouse', 'mousetrap', 'moving van', 'muzzle', 'nail', 'neck brace', 'necklace', 'nipple', 'notebook, notebook computer', 'obelisk', 'oboe, hautboy, hautbois', 'ocarina, sweet potato', 'odometer, hodometer, mileometer, milometer', 'oil filter', 'organ, pipe organ', 'oscilloscope, scope, cathode-ray oscilloscope, CRO', 'overskirt', 'oxcart', 'oxygen mask', 'packet', 'paddle, boat paddle', 'paddlewheel, paddle wheel', 'padlock', 'paintbrush', "pajama, pyjama, pj's, jammies", 'palace', 'panpipe, pandean pipe, syrinx', 'paper towel', 'parachute, chute', 'parallel bars, bars', 'park bench', 'parking meter', 'passenger car, coach, carriage', 'patio, terrace', 'pay-phone, pay-station', 'pedestal, plinth, footstall', 'pencil box, pencil case', 'pencil sharpener', 'perfume, essence', 'Petri dish', 'photocopier', 'pick, plectrum, plectron', 'pickelhaube', 'picket fence, paling', 'pickup, pickup truck', 'pier', 'piggy bank, penny bank', 'pill bottle', 'pillow', 'ping-pong ball', 'pinwheel', 'pirate, pirate ship', 'pitcher, ewer', "plane, carpenter's plane, woodworking plane", 'planetarium', 'plastic bag', 'plate rack', 'plow, plough', "plunger, plumber's helper", 'Polaroid camera, Polaroid Land camera', 'pole', 'police van, police wagon, paddy wagon, patrol wagon, wagon, black Maria', 'poncho', 'pool table, billiard table, snooker table', 'pop bottle, soda bottle', 'pot, flowerpot', "potter's wheel", 'power drill', 'prayer rug, prayer mat', 'printer', 'prison, prison house', 'projectile, missile', 'projector', 'puck, hockey puck', 'punching bag, punch bag, punching ball, punchball', 'purse', 'quill, quill pen', 'quilt, comforter, comfort, puff', 'racer, race car, racing car', 'racket, racquet', 'radiator', 'radio, wireless', 'radio telescope, radio reflector', 'rain barrel', 'recreational vehicle, RV, R.V.', 'reel', 'reflex camera', 'refrigerator, icebox', 'remote control, remote', 'restaurant, eating house, eating place, eatery', 'revolver, six-gun, six-shooter', 'rifle', 'rocking chair, rocker', 'rotisserie', 'rubber eraser, rubber, pencil eraser', 'rugby ball', 'rule, ruler', 'running shoe', 'safe', 'safety pin', 'saltshaker, salt shaker', 'sandal', 'sarong', 'sax, saxophone', 'scabbard', 'scale, weighing machine', 'school bus', 'schooner', 'scoreboard', 'screen, CRT screen', 'screw', 'screwdriver', 'seat belt, seatbelt', 'sewing machine', 'shield, buckler', 'shoe shop, shoe-shop, shoe store', 'shoji', 'shopping basket', 'shopping cart', 'shovel', 'shower cap', 'shower curtain', 'ski', 'ski mask', 'sleeping bag', 'slide rule, slipstick', 'sliding door', 'slot, one-armed bandit', 'snorkel', 'snowmobile', 'snowplow, snowplough', 'soap dispenser', 'soccer ball', 'sock', 'solar dish, solar collector, solar furnace', 'sombrero', 'soup bowl', 'space bar', 'space heater', 'space shuttle', 'spatula', 'speedboat', "spider web, spider's web", 'spindle', 'sports car, sport car', 'spotlight, spot', 'stage', 'steam locomotive', 'steel arch bridge', 'steel drum', 'stethoscope', 'stole', 'stone wall', 'stopwatch, stop watch', 'stove', 'strainer', 'streetcar, tram, tramcar, trolley, trolley car', 'stretcher', 'studio couch, day bed', 'stupa, tope', 'submarine, pigboat, sub, U-boat', 'suit, suit of clothes', 'sundial', 'sunglass', 'sunglasses, dark glasses, shades', 'sunscreen, sunblock, sun blocker', 'suspension bridge', 'swab, swob, mop', 'sweatshirt', 'swimming trunks, bathing trunks', 'swing', 'switch, electric switch, electrical switch', 'syringe', 'table lamp', 'tank, army tank, armored combat vehicle, armoured combat vehicle', 'tape player', 'teapot', 'teddy, teddy bear', 'television, television system', 'tennis ball', 'thatch, thatched roof', 'theater curtain, theatre curtain', 'thimble', 'thresher, thrasher, threshing machine', 'throne', 'tile roof', 'toaster', 'tobacco shop, tobacconist shop, tobacconist', 'toilet seat', 'torch', 'totem pole', 'tow truck, tow car, wrecker', 'toyshop', 'tractor', 'trailer truck, tractor trailer, trucking rig, rig, articulated lorry, semi', 'tray', 'trench coat', 'tricycle, trike, velocipede', 'trimaran', 'tripod', 'triumphal arch', 'trolleybus, trolley coach, trackless trolley', 'trombone', 'tub, vat', 'turnstile', 'typewriter keyboard', 'umbrella', 'unicycle, monocycle', 'upright, upright piano', 'vacuum, vacuum cleaner', 'vase', 'vault', 'velvet', 'vending machine', 'vestment', 'viaduct', 'violin, fiddle', 'volleyball', 'waffle iron', 'wall clock', 'wallet, billfold, notecase, pocketbook', 'wardrobe, closet, press', 'warplane, military plane', 'washbasin, handbasin, washbowl, lavabo, wash-hand basin', 'washer, automatic washer, washing machine', 'water bottle', 'water jug', 'water tower', 'whiskey jug', 'whistle', 'wig', 'window screen', 'window shade', 'Windsor tie', 'wine bottle', 'wing', 'wok', 'wooden spoon', 'wool, woolen, woollen', 'worm fence, snake fence, snake-rail fence, Virginia fence', 'wreck', 'yawl', 'yurt', 'web site, website, internet site, site', 'comic book', 'crossword puzzle, crossword', 'street sign', 'traffic light, traffic signal, stoplight', 'book jacket, dust cover, dust jacket, dust wrapper', 'menu', 'plate', 'guacamole', 'consomme', 'hot pot, hotpot', 'trifle', 'ice cream, icecream', 'ice lolly, lolly, lollipop, popsicle', 'French loaf', 'bagel, beigel', 'pretzel', 'cheeseburger', 'hotdog, hot dog, red hot', 'mashed potato', 'head cabbage', 'broccoli', 'cauliflower', 'zucchini, courgette', 'spaghetti squash', 'acorn squash', 'butternut squash', 'cucumber, cuke', 'artichoke, globe artichoke', 'bell pepper', 'cardoon', 'mushroom', 'Granny Smith', 'strawberry', 'orange', 'lemon', 'fig', 'pineapple, ananas', 'banana', 'jackfruit, jak, jack', 'custard apple', 'pomegranate', 'hay', 'carbonara', 'chocolate sauce, chocolate syrup', 'dough', 'meat loaf, meatloaf', 'pizza, pizza pie', 'potpie', 'burrito', 'red wine', 'espresso', 'cup', 'eggnog', 'alp', 'bubble', 'cliff, drop, drop-off', 'coral reef', 'geyser', 'lakeside, lakeshore', 'promontory, headland, head, foreland', 'sandbar, sand bar', 'seashore, coast, seacoast, sea-coast', 'valley, vale', 'volcano', 'ballplayer, baseball player', 'groom, bridegroom', 'scuba diver', 'rapeseed', 'daisy', "yellow lady's slipper, yellow lady-slipper, Cypripedium calceolus, Cypripedium parviflorum", 'corn', 'acorn', 'hip, rose hip, rosehip', 'buckeye, horse chestnut, conker', 'coral fungus', 'agaric', 'gyromitra', 'stinkhorn, carrion fungus', 'earthstar', 'hen-of-the-woods, hen of the woods, Polyporus frondosus, Grifola frondosa', 'bolete', 'ear, spike, capitulum', 'toilet tissue, toilet paper, bathroom tissue']
+  image_size: 224 # 224 for CLIP
+  image_plot_scaling: 2 # 4 for 232 num_channels, 64 for 3000
+  channels: 3
+  logit_transform: false
+  uniform_dequantization: false
+  gaussian_dequantization: false
+  random_flip: false
+  rescaled: false
+  num_workers: 4
+  tinyImages:
+    augm_type: 'none'
+    cutout_window: 16
+    out_size: 32
+    exclude_cifar: true
+    exclude_cifar10_1: true
+    offset: 0
+
+
+model:
+  sigma_begin: 1 # 50
+  num_classes: 1086 #1086 #500  # 232 for non-RATIO mode, 3000 otherwise
+  ema: true
+  ema_rate: 0.9999
+  spec_norm: false
+  sigma_dist: geometric
+  sigma_end: 0.01
+  normalization: InstanceNorm++
+  nonlinearity: elu
+  ngf: 128
+  unet: true
+
+optim:
+  weight_decay: 0.000
+  optimizer: "Adam"
+  lr: 0.0001
+  beta1: 0.9
+  beta2: 0.999
+  amsgrad: false
+  eps: 0.00000001
+utils:
+  device_ids: [0]
+evaluation:
+  evaluation_type: 'all_samples_eval' # latex_benchmark, all_samples_eval
+  base_folder: 'ACSM/slurm_start_files/exp/logits_evolution/image_samples' #'ACSM/exp/logits_evolution_evaluate/image_samples_IN1000'
+  pattern_folder: 'Appendix_plots_*' #'BENCH*' #'FAILURE*apgd*' # # benchmark/ablation : BENCH*, apgd : FAILURE*
+  ids: [94] # benchmark/ablation : [37, 72, 94], apgd: [92]
+  #[9,37,51,61,66,72,80,85,94] # the best - [5, 11, 20, 56], random out of 100 - [32, 62, 67, 36]
+

diffusion_ckpts/256x256_diffusion_uncond.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a37c32fffd316cd494cf3f35b339936debdc1576dad13fe57c42399a5dbc78b1
+size 2211383297

guided_diffusion/.gitignore

@@ -0,0 +1,3 @@
+.DS_Store
+__pycache__/
+

guided_diffusion/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2021 OpenAI
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

guided_diffusion/README.md

@@ -0,0 +1,176 @@
+# guided-diffusion
+
+This is the codebase for [Diffusion Models Beat GANS on Image Synthesis](http://arxiv.org/abs/2105.05233).
+
+This repository is based on [openai/improved-diffusion](https://github.com/openai/improved-diffusion), with modifications for classifier conditioning and architecture improvements.
+
+# Download pre-trained models
+
+We have released checkpoints for the main models in the paper. Before using these models, please review the corresponding [model card](model-card.md) to understand the intended use and limitations of these models.
+
+Here are the download links for each model checkpoint:
+
+ * 64x64 classifier: [64x64_classifier.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/64x64_classifier.pt)
+ * 64x64 diffusion: [64x64_diffusion.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/64x64_diffusion.pt)
+ * 128x128 classifier: [128x128_classifier.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/128x128_classifier.pt)
+ * 128x128 diffusion: [128x128_diffusion.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/128x128_diffusion.pt)
+ * 256x256 classifier: [256x256_classifier.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/256x256_classifier.pt)
+ * 256x256 diffusion: [256x256_diffusion.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/256x256_diffusion.pt)
+ * 256x256 diffusion (not class conditional): [256x256_diffusion_uncond.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/256x256_diffusion_uncond.pt)
+ * 512x512 classifier: [512x512_classifier.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/512x512_classifier.pt)
+ * 512x512 diffusion: [512x512_diffusion.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/512x512_diffusion.pt)
+ * 64x64 -> 256x256 upsampler: [64_256_upsampler.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/64_256_upsampler.pt)
+ * 128x128 -> 512x512 upsampler: [128_512_upsampler.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/128_512_upsampler.pt)
+ * LSUN bedroom: [lsun_bedroom.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/lsun_bedroom.pt)
+ * LSUN cat: [lsun_cat.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/lsun_cat.pt)
+ * LSUN horse: [lsun_horse.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/lsun_horse.pt)
+ * LSUN horse (no dropout): [lsun_horse_nodropout.pt](https://openaipublic.blob.core.windows.net/diffusion/jul-2021/lsun_horse_nodropout.pt)
+
+# Sampling from pre-trained models
+
+To sample from these models, you can use the `classifier_sample.py`, `image_sample.py`, and `super_res_sample.py` scripts.
+Here, we provide flags for sampling from all of these models.
+We assume that you have downloaded the relevant model checkpoints into a folder called `models/`.
+
+For these examples, we will generate 100 samples with batch size 4. Feel free to change these values.
+
+```
+SAMPLE_FLAGS="--batch_size 4 --num_samples 100 --timestep_respacing 250"
+```
+
+## Classifier guidance
+
+Note for these sampling runs that you can set `--classifier_scale 0` to sample from the base diffusion model.
+You may also use the `image_sample.py` script instead of `classifier_sample.py` in that case.
+
+ * 64x64 model:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --diffusion_steps 1000 --dropout 0.1 --image_size 64 --learn_sigma True --noise_schedule cosine --num_channels 192 --num_head_channels 64 --num_res_blocks 3 --resblock_updown True --use_new_attention_order True --use_fp16 True --use_scale_shift_norm True"
+python classifier_sample.py $MODEL_FLAGS --classifier_scale 1.0 --classifier_path models/64x64_classifier.pt --model_path models/64x64_diffusion.pt $SAMPLE_FLAGS
+```
+
+ * 128x128 model:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --diffusion_steps 1000 --image_size 128 --learn_sigma True --noise_schedule linear --num_channels 256 --num_heads 4 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python classifier_sample.py $MODEL_FLAGS --classifier_scale 0.5 --classifier_path models/128x128_classifier.pt --model_path models/128x128_diffusion.pt $SAMPLE_FLAGS
+```
+
+ * 256x256 model:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --diffusion_steps 1000 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python classifier_sample.py $MODEL_FLAGS --classifier_scale 1.0 --classifier_path models/256x256_classifier.pt --model_path models/256x256_diffusion.pt $SAMPLE_FLAGS
+```
+
+ * 256x256 model (unconditional):
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python classifier_sample.py $MODEL_FLAGS --classifier_scale 10.0 --classifier_path models/256x256_classifier.pt --model_path models/256x256_diffusion.pt $SAMPLE_FLAGS
+```
+
+ * 512x512 model:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --diffusion_steps 1000 --image_size 512 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 False --use_scale_shift_norm True"
+python classifier_sample.py $MODEL_FLAGS --classifier_scale 4.0 --classifier_path models/512x512_classifier.pt --model_path models/512x512_diffusion.pt $SAMPLE_FLAGS
+```
+
+## Upsampling
+
+For these runs, we assume you have some base samples in a file `64_samples.npz` or `128_samples.npz` for the two respective models.
+
+ * 64 -> 256:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --diffusion_steps 1000 --large_size 256  --small_size 64 --learn_sigma True --noise_schedule linear --num_channels 192 --num_heads 4 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python super_res_sample.py $MODEL_FLAGS --model_path models/64_256_upsampler.pt --base_samples 64_samples.npz $SAMPLE_FLAGS
+```
+
+ * 128 -> 512:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16 --class_cond True --diffusion_steps 1000 --large_size 512 --small_size 128 --learn_sigma True --noise_schedule linear --num_channels 192 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python super_res_sample.py $MODEL_FLAGS --model_path models/128_512_upsampler.pt $SAMPLE_FLAGS --base_samples 128_samples.npz
+```
+
+## LSUN models
+
+These models are class-unconditional and correspond to a single LSUN class. Here, we show how to sample from `lsun_bedroom.pt`, but the other two LSUN checkpoints should work as well:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --dropout 0.1 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python image_sample.py $MODEL_FLAGS --model_path models/lsun_bedroom.pt $SAMPLE_FLAGS
+```
+
+You can sample from `lsun_horse_nodropout.pt` by changing the dropout flag:
+
+```
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond False --diffusion_steps 1000 --dropout 0.0 --image_size 256 --learn_sigma True --noise_schedule linear --num_channels 256 --num_head_channels 64 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+python image_sample.py $MODEL_FLAGS --model_path models/lsun_horse_nodropout.pt $SAMPLE_FLAGS
+```
+
+Note that for these models, the best samples result from using 1000 timesteps:
+
+```
+SAMPLE_FLAGS="--batch_size 4 --num_samples 100 --timestep_respacing 1000"
+```
+
+# Results
+
+This table summarizes our ImageNet results for pure guided diffusion models:
+
+| Dataset          | FID  | Precision | Recall |
+|------------------|------|-----------|--------|
+| ImageNet 64x64   | 2.07 | 0.74      | 0.63   |
+| ImageNet 128x128 | 2.97 | 0.78      | 0.59   |
+| ImageNet 256x256 | 4.59 | 0.82      | 0.52   |
+| ImageNet 512x512 | 7.72 | 0.87      | 0.42   |
+
+This table shows the best results for high resolutions when using upsampling and guidance together:
+
+| Dataset          | FID  | Precision | Recall |
+|------------------|------|-----------|--------|
+| ImageNet 256x256 | 3.94 | 0.83      | 0.53   |
+| ImageNet 512x512 | 3.85 | 0.84      | 0.53   |
+
+Finally, here are the unguided results on individual LSUN classes:
+
+| Dataset      | FID  | Precision | Recall |
+|--------------|------|-----------|--------|
+| LSUN Bedroom | 1.90 | 0.66      | 0.51   |
+| LSUN Cat     | 5.57 | 0.63      | 0.52   |
+| LSUN Horse   | 2.57 | 0.71      | 0.55   |
+
+# Training models
+
+Training diffusion models is described in the [parent repository](https://github.com/openai/improved-diffusion). Training a classifier is similar. We assume you have put training hyperparameters into a `TRAIN_FLAGS` variable, and classifier hyperparameters into a `CLASSIFIER_FLAGS` variable. Then you can run:
+
+```
+mpiexec -n N python scripts/classifier_train.py --data_dir path/to/imagenet $TRAIN_FLAGS $CLASSIFIER_FLAGS
+```
+
+Make sure to divide the batch size in `TRAIN_FLAGS` by the number of MPI processes you are using.
+
+Here are flags for training the 128x128 classifier. You can modify these for training classifiers at other resolutions:
+
+```sh
+TRAIN_FLAGS="--iterations 300000 --anneal_lr True --batch_size 256 --lr 3e-4 --save_interval 10000 --weight_decay 0.05"
+CLASSIFIER_FLAGS="--image_size 128 --classifier_attention_resolutions 32,16,8 --classifier_depth 2 --classifier_width 128 --classifier_pool attention --classifier_resblock_updown True --classifier_use_scale_shift_norm True"
+```
+
+For sampling from a 128x128 classifier-guided model, 25 step DDIM:
+
+```sh
+MODEL_FLAGS="--attention_resolutions 32,16,8 --class_cond True --image_size 128 --learn_sigma True --num_channels 256 --num_heads 4 --num_res_blocks 2 --resblock_updown True --use_fp16 True --use_scale_shift_norm True"
+CLASSIFIER_FLAGS="--image_size 128 --classifier_attention_resolutions 32,16,8 --classifier_depth 2 --classifier_width 128 --classifier_pool attention --classifier_resblock_updown True --classifier_use_scale_shift_norm True --classifier_scale 1.0 --classifier_use_fp16 True"
+SAMPLE_FLAGS="--batch_size 4 --num_samples 50000 --timestep_respacing ddim25 --use_ddim True"
+mpiexec -n N python scripts/classifier_sample.py \
+    --model_path /path/to/model.pt \
+    --classifier_path path/to/classifier.pt \
+    $MODEL_FLAGS $CLASSIFIER_FLAGS $SAMPLE_FLAGS
+```
+
+To sample for 250 timesteps without DDIM, replace `--timestep_respacing ddim25` to `--timestep_respacing 250`, and replace `--use_ddim True` with `--use_ddim False`.

guided_diffusion/dist_util.py

@@ -0,0 +1,93 @@
+"""
+Helpers for distributed training.
+"""
+
+import io
+import os
+import socket
+
+import blobfile as bf
+from mpi4py import MPI
+import torch as th
+import torch.distributed as dist
+
+# Change this to reflect your cluster layout.
+# The GPU for a given rank is (rank % GPUS_PER_NODE).
+GPUS_PER_NODE = 8
+
+SETUP_RETRY_COUNT = 3
+
+
+def setup_dist():
+    """
+    Setup a distributed process group.
+    """
+    if dist.is_initialized():
+        return
+    os.environ["CUDA_VISIBLE_DEVICES"] = f"{MPI.COMM_WORLD.Get_rank() % GPUS_PER_NODE}"
+
+    comm = MPI.COMM_WORLD
+    backend = "gloo" if not th.cuda.is_available() else "nccl"
+
+    if backend == "gloo":
+        hostname = "localhost"
+    else:
+        hostname = socket.gethostbyname(socket.getfqdn())
+    os.environ["MASTER_ADDR"] = comm.bcast(hostname, root=0)
+    os.environ["RANK"] = str(comm.rank)
+    os.environ["WORLD_SIZE"] = str(comm.size)
+
+    port = comm.bcast(_find_free_port(), root=0)
+    os.environ["MASTER_PORT"] = str(port)
+    dist.init_process_group(backend=backend, init_method="env://")
+
+
+def dev():
+    """
+    Get the device to use for torch.distributed.
+    """
+    if th.cuda.is_available():
+        return th.device(f"cuda")
+    return th.device("cpu")
+
+
+def load_state_dict(path, **kwargs):
+    """
+    Load a PyTorch file without redundant fetches across MPI ranks.
+    """
+    chunk_size = 2 ** 30  # MPI has a relatively small size limit
+    if MPI.COMM_WORLD.Get_rank() == 0:
+        with bf.BlobFile(path, "rb") as f:
+            data = f.read()
+        num_chunks = len(data) // chunk_size
+        if len(data) % chunk_size:
+            num_chunks += 1
+        MPI.COMM_WORLD.bcast(num_chunks)
+        for i in range(0, len(data), chunk_size):
+            MPI.COMM_WORLD.bcast(data[i : i + chunk_size])
+    else:
+        num_chunks = MPI.COMM_WORLD.bcast(None)
+        data = bytes()
+        for _ in range(num_chunks):
+            data += MPI.COMM_WORLD.bcast(None)
+
+    return th.load(io.BytesIO(data), **kwargs)
+
+
+def sync_params(params):
+    """
+    Synchronize a sequence of Tensors across ranks from rank 0.
+    """
+    for p in params:
+        with th.no_grad():
+            dist.broadcast(p, 0)
+
+
+def _find_free_port():
+    try:
+        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        s.bind(("", 0))
+        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        return s.getsockname()[1]
+    finally:
+        s.close()

guided_diffusion/guided_diffusion/__init__.py

@@ -0,0 +1,3 @@
+"""
+Codebase for "Improved Denoising Diffusion Probabilistic Models".
+"""

guided_diffusion/guided_diffusion/dist_util.py

@@ -0,0 +1,93 @@
+"""
+Helpers for distributed training.
+"""
+
+import io
+import os
+import socket
+
+import blobfile as bf
+from mpi4py import MPI
+import torch as th
+import torch.distributed as dist
+
+# Change this to reflect your cluster layout.
+# The GPU for a given rank is (rank % GPUS_PER_NODE).
+GPUS_PER_NODE = 8
+
+SETUP_RETRY_COUNT = 3
+
+
+def setup_dist():
+    """
+    Setup a distributed process group.
+    """
+    if dist.is_initialized():
+        return
+    os.environ["CUDA_VISIBLE_DEVICES"] = f"{MPI.COMM_WORLD.Get_rank() % GPUS_PER_NODE}"
+
+    comm = MPI.COMM_WORLD
+    backend = "gloo" if not th.cuda.is_available() else "nccl"
+
+    if backend == "gloo":
+        hostname = "localhost"
+    else:
+        hostname = socket.gethostbyname(socket.getfqdn())
+    os.environ["MASTER_ADDR"] = comm.bcast(hostname, root=0)
+    os.environ["RANK"] = str(comm.rank)
+    os.environ["WORLD_SIZE"] = str(comm.size)
+
+    port = comm.bcast(_find_free_port(), root=0)
+    os.environ["MASTER_PORT"] = str(port)
+    dist.init_process_group(backend=backend, init_method="env://")
+
+
+def dev():
+    """
+    Get the device to use for torch.distributed.
+    """
+    if th.cuda.is_available():
+        return th.device(f"cuda")
+    return th.device("cpu")
+
+
+def load_state_dict(path, **kwargs):
+    """
+    Load a PyTorch file without redundant fetches across MPI ranks.
+    """
+    chunk_size = 2 ** 30  # MPI has a relatively small size limit
+    if MPI.COMM_WORLD.Get_rank() == 0:
+        with bf.BlobFile(path, "rb") as f:
+            data = f.read()
+        num_chunks = len(data) // chunk_size
+        if len(data) % chunk_size:
+            num_chunks += 1
+        MPI.COMM_WORLD.bcast(num_chunks)
+        for i in range(0, len(data), chunk_size):
+            MPI.COMM_WORLD.bcast(data[i : i + chunk_size])
+    else:
+        num_chunks = MPI.COMM_WORLD.bcast(None)
+        data = bytes()
+        for _ in range(num_chunks):
+            data += MPI.COMM_WORLD.bcast(None)
+
+    return th.load(io.BytesIO(data), **kwargs)
+
+
+def sync_params(params):
+    """
+    Synchronize a sequence of Tensors across ranks from rank 0.
+    """
+    for p in params:
+        with th.no_grad():
+            dist.broadcast(p, 0)
+
+
+def _find_free_port():
+    try:
+        s = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
+        s.bind(("", 0))
+        s.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
+        return s.getsockname()[1]
+    finally:
+        s.close()

guided_diffusion/guided_diffusion/fp16_util.py

@@ -0,0 +1,236 @@
+"""
+Helpers to train with 16-bit precision.
+"""
+
+import numpy as np
+import torch as th
+import torch.nn as nn
+from torch._utils import _flatten_dense_tensors, _unflatten_dense_tensors
+
+from . import logger
+
+INITIAL_LOG_LOSS_SCALE = 20.0
+
+
+def convert_module_to_f16(l):
+    """
+    Convert primitive modules to float16.
+    """
+    if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):
+        l.weight.data = l.weight.data.half()
+        if l.bias is not None:
+            l.bias.data = l.bias.data.half()
+
+
+def convert_module_to_f32(l):
+    """
+    Convert primitive modules to float32, undoing convert_module_to_f16().
+    """
+    if isinstance(l, (nn.Conv1d, nn.Conv2d, nn.Conv3d)):
+        l.weight.data = l.weight.data.float()
+        if l.bias is not None:
+            l.bias.data = l.bias.data.float()
+
+
+def make_master_params(param_groups_and_shapes):
+    """
+    Copy model parameters into a (differently-shaped) list of full-precision
+    parameters.
+    """
+    master_params = []
+    for param_group, shape in param_groups_and_shapes:
+        master_param = nn.Parameter(
+            _flatten_dense_tensors(
+                [param.detach().float() for (_, param) in param_group]
+            ).view(shape)
+        )
+        master_param.requires_grad = True
+        master_params.append(master_param)
+    return master_params
+
+
+def model_grads_to_master_grads(param_groups_and_shapes, master_params):
+    """
+    Copy the gradients from the model parameters into the master parameters
+    from make_master_params().
+    """
+    for master_param, (param_group, shape) in zip(
+        master_params, param_groups_and_shapes
+    ):
+        master_param.grad = _flatten_dense_tensors(
+            [param_grad_or_zeros(param) for (_, param) in param_group]
+        ).view(shape)
+
+
+def master_params_to_model_params(param_groups_and_shapes, master_params):
+    """
+    Copy the master parameter data back into the model parameters.
+    """
+    # Without copying to a list, if a generator is passed, this will
+    # silently not copy any parameters.
+    for master_param, (param_group, _) in zip(master_params, param_groups_and_shapes):
+        for (_, param), unflat_master_param in zip(
+            param_group, unflatten_master_params(param_group, master_param.view(-1))
+        ):
+            param.detach().copy_(unflat_master_param)
+
+
+def unflatten_master_params(param_group, master_param):
+    return _unflatten_dense_tensors(master_param, [param for (_, param) in param_group])
+
+
+def get_param_groups_and_shapes(named_model_params):
+    named_model_params = list(named_model_params)
+    scalar_vector_named_params = (
+        [(n, p) for (n, p) in named_model_params if p.ndim <= 1],
+        (-1),
+    )
+    matrix_named_params = (
+        [(n, p) for (n, p) in named_model_params if p.ndim > 1],
+        (1, -1),
+    )
+    return [scalar_vector_named_params, matrix_named_params]
+
+
+def master_params_to_state_dict(
+    model, param_groups_and_shapes, master_params, use_fp16
+):
+    if use_fp16:
+        state_dict = model.state_dict()
+        for master_param, (param_group, _) in zip(
+            master_params, param_groups_and_shapes
+        ):
+            for (name, _), unflat_master_param in zip(
+                param_group, unflatten_master_params(param_group, master_param.view(-1))
+            ):
+                assert name in state_dict
+                state_dict[name] = unflat_master_param
+    else:
+        state_dict = model.state_dict()
+        for i, (name, _value) in enumerate(model.named_parameters()):
+            assert name in state_dict
+            state_dict[name] = master_params[i]
+    return state_dict
+
+
+def state_dict_to_master_params(model, state_dict, use_fp16):
+    if use_fp16:
+        named_model_params = [
+            (name, state_dict[name]) for name, _ in model.named_parameters()
+        ]
+        param_groups_and_shapes = get_param_groups_and_shapes(named_model_params)
+        master_params = make_master_params(param_groups_and_shapes)
+    else:
+        master_params = [state_dict[name] for name, _ in model.named_parameters()]
+    return master_params
+
+
+def zero_master_grads(master_params):
+    for param in master_params:
+        param.grad = None
+
+
+def zero_grad(model_params):
+    for param in model_params:
+        # Taken from https://pytorch.org/docs/stable/_modules/torch/optim/optimizer.html#Optimizer.add_param_group
+        if param.grad is not None:
+            param.grad.detach_()
+            param.grad.zero_()
+
+
+def param_grad_or_zeros(param):
+    if param.grad is not None:
+        return param.grad.data.detach()
+    else:
+        return th.zeros_like(param)
+
+
+class MixedPrecisionTrainer:
+    def __init__(
+        self,
+        *,
+        model,
+        use_fp16=False,
+        fp16_scale_growth=1e-3,
+        initial_lg_loss_scale=INITIAL_LOG_LOSS_SCALE,
+    ):
+        self.model = model
+        self.use_fp16 = use_fp16
+        self.fp16_scale_growth = fp16_scale_growth
+
+        self.model_params = list(self.model.parameters())
+        self.master_params = self.model_params
+        self.param_groups_and_shapes = None
+        self.lg_loss_scale = initial_lg_loss_scale
+
+        if self.use_fp16:
+            self.param_groups_and_shapes = get_param_groups_and_shapes(
+                self.model.named_parameters()
+            )
+            self.master_params = make_master_params(self.param_groups_and_shapes)
+            self.model.convert_to_fp16()
+
+    def zero_grad(self):
+        zero_grad(self.model_params)
+
+    def backward(self, loss: th.Tensor):
+        if self.use_fp16:
+            loss_scale = 2 ** self.lg_loss_scale
+            (loss * loss_scale).backward()
+        else:
+            loss.backward()
+
+    def optimize(self, opt: th.optim.Optimizer):
+        if self.use_fp16:
+            return self._optimize_fp16(opt)
+        else:
+            return self._optimize_normal(opt)
+
+    def _optimize_fp16(self, opt: th.optim.Optimizer):
+        logger.logkv_mean("lg_loss_scale", self.lg_loss_scale)
+        model_grads_to_master_grads(self.param_groups_and_shapes, self.master_params)
+        grad_norm, param_norm = self._compute_norms(grad_scale=2 ** self.lg_loss_scale)
+        if check_overflow(grad_norm):
+            self.lg_loss_scale -= 1
+            logger.log(f"Found NaN, decreased lg_loss_scale to {self.lg_loss_scale}")
+            zero_master_grads(self.master_params)
+            return False
+
+        logger.logkv_mean("grad_norm", grad_norm)
+        logger.logkv_mean("param_norm", param_norm)
+
+        self.master_params[0].grad.mul_(1.0 / (2 ** self.lg_loss_scale))
+        opt.step()
+        zero_master_grads(self.master_params)
+        master_params_to_model_params(self.param_groups_and_shapes, self.master_params)
+        self.lg_loss_scale += self.fp16_scale_growth
+        return True
+
+    def _optimize_normal(self, opt: th.optim.Optimizer):
+        grad_norm, param_norm = self._compute_norms()
+        logger.logkv_mean("grad_norm", grad_norm)
+        logger.logkv_mean("param_norm", param_norm)
+        opt.step()
+        return True
+
+    def _compute_norms(self, grad_scale=1.0):
+        grad_norm = 0.0
+        param_norm = 0.0
+        for p in self.master_params:
+            with th.no_grad():
+                param_norm += th.norm(p, p=2, dtype=th.float32).item() ** 2
+                if p.grad is not None:
+                    grad_norm += th.norm(p.grad, p=2, dtype=th.float32).item() ** 2
+        return np.sqrt(grad_norm) / grad_scale, np.sqrt(param_norm)
+
+    def master_params_to_state_dict(self, master_params):
+        return master_params_to_state_dict(
+            self.model, self.param_groups_and_shapes, master_params, self.use_fp16
+        )
+
+    def state_dict_to_master_params(self, state_dict):
+        return state_dict_to_master_params(self.model, state_dict, self.use_fp16)
+
+
+def check_overflow(value):
+    return (value == float("inf")) or (value == -float("inf")) or (value != value)

guided_diffusion/guided_diffusion/gaussian_diffusion.py

@@ -0,0 +1,1252 @@
+"""
+This code started out as a PyTorch port of Ho et al's diffusion models:
+https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/diffusion_utils_2.py
+
+Docstrings have been added, as well as DDIM sampling and a new collection of beta schedules.
+"""
+
+import enum
+import math
+
+import numpy as np
+import torch # TODO: I know this is messy, should clean up later
+import torch as th
+import random
+import matplotlib.pyplot as plt
+
+from .nn import mean_flat
+from .losses import normal_kl, discretized_gaussian_log_likelihood
+from tqdm import tqdm
+
+
+def compute_alpha(beta, t):
+    if isinstance(beta, np.ndarray):
+        beta = torch.Tensor(beta)
+    beta = torch.cat([torch.zeros(1).to(beta.device), beta], dim=0)
+    t = t.to(beta.device)
+    a = (1 - beta).cumprod(dim=0).index_select(0, t + 1).view(-1, 1, 1, 1)
+    return a.to(beta.device)
+
+
+def show_tensor(img): # (3, 256, 256)
+    img = img.detach().cpu().numpy()
+    img = np.transpose(img, (1,2,0))
+    plt.imshow(img)
+    plt.show()
+
+
+def L1_projection(x2, y2, eps1):
+    '''
+    x2: center of the L1 ball (bs x input_dim)
+    y2: current perturbation (x2 + y2 is the point to be projected)
+    eps1: radius of the L1 ball
+
+    output: delta s.th. ||y2 + delta||_1 <= eps1
+    and 0 <= x2 + y2 + delta <= 1
+    '''
+
+    x = x2.clone().float().view(x2.shape[0], -1)
+    y = y2.clone().float().view(y2.shape[0], -1)
+    sigma = y.clone().sign()
+    u = th.min(1 - x - y, x + y)
+    # u = th.min(u, epsinf - th.clone(y).abs())
+    u = th.min(th.zeros_like(y), u)
+    l = -th.clone(y).abs()
+    d = u.clone()
+
+    bs, indbs = th.sort(-th.cat((u, l), 1), dim=1)
+    bs2 = th.cat((bs[:, 1:], th.zeros(bs.shape[0], 1).to(bs.device)), 1)
+
+    inu = 2 * (indbs < u.shape[1]).float() - 1
+    size1 = inu.cumsum(dim=1)
+
+    s1 = -u.sum(dim=1)
+
+    c = eps1 - y.clone().abs().sum(dim=1)
+    c5 = s1 + c < 0
+    c2 = c5.nonzero().squeeze(1)
+
+    s = s1.unsqueeze(-1) + th.cumsum((bs2 - bs) * size1, dim=1)
+
+    if c2.nelement != 0:
+
+        lb = th.zeros_like(c2).float()
+        ub = th.ones_like(lb) * (bs.shape[1] - 1)
+
+        # print(c2.shape, lb.shape)
+
+        nitermax = th.ceil(th.log2(th.tensor(bs.shape[1]).float()))
+        counter2 = th.zeros_like(lb).long()
+        counter = 0
+
+        while counter < nitermax:
+            counter4 = th.floor((lb + ub) / 2.)
+            counter2 = counter4.type(th.LongTensor)
+
+            c8 = s[c2, counter2] + c[c2] < 0
+            ind3 = c8.nonzero().squeeze(1)
+            ind32 = (~c8).nonzero().squeeze(1)
+            if ind3.nelement != 0:
+                lb[ind3] = counter4[ind3]
+            if ind32.nelement != 0:
+                ub[ind32] = counter4[ind32]
+
+            counter += 1
+
+        lb2 = lb.long()
+        alpha = (-s[c2, lb2] - c[c2]) / size1[c2, lb2 + 1] + bs2[c2, lb2]
+        d[c2] = -th.min(th.max(-u[c2], alpha.unsqueeze(-1)), -l[c2])
+
+    return (sigma * d).view(x2.shape)
+
+def project_perturbation(perturbation, eps, p, center=None):
+    if p in ['inf', 'linf', 'Linf']:
+        pert_normalized = th.clamp(perturbation, -eps, eps)
+        return pert_normalized
+    elif p in [2, 2.0, 'l2', 'L2', '2']:
+        print('l2 renorm')
+        pert_normalized = th.renorm(perturbation, p=2, dim=0, maxnorm=eps)
+        return pert_normalized
+    elif p in [1, 1.0, 'l1', 'L1', '1']:
+        ##pert_normalized = project_onto_l1_ball(perturbation, eps)
+        ##return pert_normalized
+        pert_normalized = L1_projection(center, perturbation, eps)
+        return perturbation + pert_normalized
+    #elif p in ['LPIPS']:
+    #    pert_normalized = project_onto_LPIPS_ball(perturbation, eps)
+    else:
+        raise NotImplementedError('Projection only supports l1, l2 and inf norm')
+
+def get_named_beta_schedule(schedule_name, num_diffusion_timesteps):
+    """
+    Get a pre-defined beta schedule for the given name.
+
+    The beta schedule library consists of beta schedules which remain similar
+    in the limit of num_diffusion_timesteps.
+    Beta schedules may be added, but should not be removed or changed once
+    they are committed to maintain backwards compatibility.
+    """
+    if schedule_name == "linear":
+        # Linear schedule from Ho et al, extended to work for any number of
+        # diffusion steps.
+        scale = 1000 / num_diffusion_timesteps
+        beta_start = scale * 0.0001
+        beta_end = scale * 0.02
+        return np.linspace(beta_start, beta_end, num_diffusion_timesteps, dtype=np.float64)
+    elif schedule_name == "cosine":
+        return betas_for_alpha_bar(
+            num_diffusion_timesteps, lambda t: math.cos((t + 0.008) / 1.008 * math.pi / 2) ** 2,
+        )
+    else:
+        raise NotImplementedError(f"unknown beta schedule: {schedule_name}")
+
+
+def betas_for_alpha_bar(num_diffusion_timesteps, alpha_bar, max_beta=0.999):
+    """
+    Create a beta schedule that discretizes the given alpha_t_bar function,
+    which defines the cumulative product of (1-beta) over time from t = [0,1].
+
+    :param num_diffusion_timesteps: the number of betas to produce.
+    :param alpha_bar: a lambda that takes an argument t from 0 to 1 and
+                      produces the cumulative product of (1-beta) up to that
+                      part of the diffusion process.
+    :param max_beta: the maximum beta to use; use values lower than 1 to
+                     prevent singularities.
+    """
+    betas = []
+    for i in range(num_diffusion_timesteps):
+        t1 = i / num_diffusion_timesteps
+        t2 = (i + 1) / num_diffusion_timesteps
+        betas.append(min(1 - alpha_bar(t2) / alpha_bar(t1), max_beta))
+    return np.array(betas)
+
+
+class ModelMeanType(enum.Enum):
+    """
+    Which type of output the model predicts.
+    """
+
+    PREVIOUS_X = enum.auto()  # the model predicts x_{t-1}
+    START_X = enum.auto()  # the model predicts x_0
+    EPSILON = enum.auto()  # the model predicts epsilon
+
+
+class ModelVarType(enum.Enum):
+    """
+    What is used as the model's output variance.
+
+    The LEARNED_RANGE option has been added to allow the model to predict
+    values between FIXED_SMALL and FIXED_LARGE, making its job easier.
+    """
+
+    LEARNED = enum.auto()
+    FIXED_SMALL = enum.auto()
+    FIXED_LARGE = enum.auto()
+    LEARNED_RANGE = enum.auto()
+
+
+class LossType(enum.Enum):
+    MSE = enum.auto()  # use raw MSE loss (and KL when learning variances)
+    RESCALED_MSE = enum.auto()  # use raw MSE loss (with RESCALED_KL when learning variances)
+    KL = enum.auto()  # use the variational lower-bound
+    RESCALED_KL = enum.auto()  # like KL, but rescale to estimate the full VLB
+
+    def is_vb(self):
+        return self == LossType.KL or self == LossType.RESCALED_KL
+
+
+class GaussianDiffusion:
+    """
+    Utilities for training and sampling diffusion models.
+
+    Ported directly from here, and then adapted over time to further experimentation.
+    https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/diffusion_utils_2.py#L42
+
+    :param betas: a 1-D numpy array of betas for each diffusion timestep,
+                  starting at T and going to 1.
+    :param model_mean_type: a ModelMeanType determining what the model outputs.
+    :param model_var_type: a ModelVarType determining how variance is output.
+    :param loss_type: a LossType determining the loss function to use.
+    :param rescale_timesteps: if True, pass floating point timesteps into the
+                              model so that they are always scaled like in the
+                              original paper (0 to 1000).
+    """
+
+    def __init__(
+        self, *, betas, model_mean_type, model_var_type, loss_type, rescale_timesteps=False,
+    ):
+        self.model_mean_type = model_mean_type
+        self.model_var_type = model_var_type
+        self.loss_type = loss_type
+        self.rescale_timesteps = rescale_timesteps
+
+        # Use float64 for accuracy.
+        betas = np.array(betas, dtype=np.float64)
+        self.betas = betas
+        assert len(betas.shape) == 1, "betas must be 1-D"
+        assert (betas > 0).all() and (betas <= 1).all()
+
+        self.num_timesteps = int(betas.shape[0])
+
+        alphas = 1.0 - betas
+        self.alphas_cumprod = np.cumprod(alphas, axis=0)
+        self.alphas_cumprod_prev = np.append(1.0, self.alphas_cumprod[:-1])
+        self.alphas_cumprod_next = np.append(self.alphas_cumprod[1:], 0.0)
+        assert self.alphas_cumprod_prev.shape == (self.num_timesteps,)
+
+        # calculations for diffusion q(x_t | x_{t-1}) and others
+        self.sqrt_alphas_cumprod = np.sqrt(self.alphas_cumprod)
+        self.sqrt_one_minus_alphas_cumprod = np.sqrt(1.0 - self.alphas_cumprod)
+        self.log_one_minus_alphas_cumprod = np.log(1.0 - self.alphas_cumprod)
+        self.sqrt_recip_alphas_cumprod = np.sqrt(1.0 / self.alphas_cumprod)
+        self.sqrt_recipm1_alphas_cumprod = np.sqrt(1.0 / self.alphas_cumprod - 1)
+ 
+        # calculations for posterior q(x_{t-1} | x_t, x_0)
+        self.posterior_variance = (
+            betas * (1.0 - self.alphas_cumprod_prev) / (1.0 - self.alphas_cumprod)
+        )
+        # log calculation clipped because the posterior variance is 0 at the
+        # beginning of the diffusion chain.
+        self.posterior_log_variance_clipped = np.log(
+            np.append(self.posterior_variance[1], self.posterior_variance[1:])
+        )
+        self.posterior_mean_coef1 = (
+            betas * np.sqrt(self.alphas_cumprod_prev) / (1.0 - self.alphas_cumprod)
+        )
+        self.posterior_mean_coef2 = (
+            (1.0 - self.alphas_cumprod_prev) * np.sqrt(alphas) / (1.0 - self.alphas_cumprod)
+        )
+
+        self.applied_noise = None
+
+    def q_mean_variance(self, x_start, t):
+        """
+        Get the distribution q(x_t | x_0).
+
+        :param x_start: the [N x C x ...] tensor of noiseless inputs.
+        :param t: the number of diffusion steps (minus 1). Here, 0 means one step.
+        :return: A tuple (mean, variance, log_variance), all of x_start's shape.
+        """
+        mean = _extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
+        variance = _extract_into_tensor(1.0 - self.alphas_cumprod, t, x_start.shape)
+        log_variance = _extract_into_tensor(self.log_one_minus_alphas_cumprod, t, x_start.shape)
+        return mean, variance, log_variance
+
+    def q_sample(self, x_start, t, noise=None):
+        """
+        Diffuse the data for a given number of diffusion steps.
+
+        In other words, sample from q(x_t | x_0).
+
+        :param x_start: the initial data batch.
+        :param t: the number of diffusion steps (minus 1). Here, 0 means one step.
+        :param noise: if specified, the split-out normal noise.
+        :return: A noisy version of x_start.
+        """
+        if noise is None:
+            noise = th.randn_like(x_start)
+        assert noise.shape == x_start.shape
+        a = _extract_into_tensor(self.sqrt_alphas_cumprod, t, x_start.shape) * x_start
+        b = _extract_into_tensor(self.sqrt_one_minus_alphas_cumprod, t, x_start.shape) * noise
+        out = a + b
+        return out
+
+    def q_posterior_mean_variance(self, x_start, x_t, t):
+        """
+        Compute the mean and variance of the diffusion posterior:
+
+            q(x_{t-1} | x_t, x_0)
+
+        """
+        assert x_start.shape == x_t.shape
+        posterior_mean = (
+            _extract_into_tensor(self.posterior_mean_coef1, t, x_t.shape) * x_start
+            + _extract_into_tensor(self.posterior_mean_coef2, t, x_t.shape) * x_t
+        )
+        posterior_variance = _extract_into_tensor(self.posterior_variance, t, x_t.shape)
+        posterior_log_variance_clipped = _extract_into_tensor(
+            self.posterior_log_variance_clipped, t, x_t.shape
+        )
+        assert (
+            posterior_mean.shape[0]
+            == posterior_variance.shape[0]
+            == posterior_log_variance_clipped.shape[0]
+            == x_start.shape[0]
+        )
+        return posterior_mean, posterior_variance, posterior_log_variance_clipped
+
+    def p_mean_variance(self, model, x, t, clip_denoised=True, denoised_fn=None, model_kwargs=None):
+        """
+        Apply the model to get p(x_{t-1} | x_t), as well as a prediction of
+        the initial x, x_0.
+
+        :param model: the model, which takes a signal and a batch of timesteps
+                      as input.
+        :param x: the [N x C x ...] tensor at time t.
+        :param t: a 1-D Tensor of timesteps.
+        :param clip_denoised: if True, clip the denoised signal into [-1, 1].
+        :param denoised_fn: if not None, a function which applies to the
+            x_start prediction before it is used to sample. Applies before
+            clip_denoised.
+        :param model_kwargs: if not None, a dict of extra keyword arguments to
+            pass to the model. This can be used for conditioning.
+        :return: a dict with the following keys:
+                 - 'mean': the model mean output.
+                 - 'variance': the model variance output.
+                 - 'log_variance': the log of 'variance'.
+                 - 'pred_xstart': the prediction for x_0.
+        """
+        if model_kwargs is None:
+            model_kwargs = {}
+
+        B, C = x.shape[:2]
+        assert t.shape == (B,)
+
+        if "middle_out" not in model_kwargs:
+            model_kwargs["middle_out"] = False
+
+        if model_kwargs["middle_out"]:
+            model_output, middle_out = model(x, self._scale_timesteps(t), **model_kwargs)
+        else:
+            model_output = model(x, self._scale_timesteps(t), **model_kwargs)
+            middle_out = None
+
+        del model_kwargs["middle_out"]
+
+
+        if self.model_var_type in [ModelVarType.LEARNED, ModelVarType.LEARNED_RANGE]:
+            assert model_output.shape == (B, C * 2, *x.shape[2:])
+            model_output, model_var_values = th.split(model_output, C, dim=1)
+            if self.model_var_type == ModelVarType.LEARNED:
+                model_log_variance = model_var_values
+                model_variance = th.exp(model_log_variance)
+            else:
+                min_log = _extract_into_tensor(self.posterior_log_variance_clipped, t, x.shape)
+                max_log = _extract_into_tensor(np.log(self.betas), t, x.shape)
+                # The model_var_values is [-1, 1] for [min_var, max_var].
+                frac = (model_var_values + 1) / 2
+                model_log_variance = frac * max_log + (1 - frac) * min_log
+                model_variance = th.exp(model_log_variance)
+        else:
+            model_variance, model_log_variance = {
+                # for fixedlarge, we set the initial (log-)variance like so
+                # to get a better decoder log likelihood.
+                ModelVarType.FIXED_LARGE: (
+                    np.append(self.posterior_variance[1], self.betas[1:]),
+                    np.log(np.append(self.posterior_variance[1], self.betas[1:])),
+                ),
+                ModelVarType.FIXED_SMALL: (
+                    self.posterior_variance,
+                    self.posterior_log_variance_clipped,
+                ),
+            }[self.model_var_type]
+            model_variance = _extract_into_tensor(model_variance, t, x.shape)
+            model_log_variance = _extract_into_tensor(model_log_variance, t, x.shape)
+
+        def process_xstart(x):
+            if denoised_fn is not None:
+                x = denoised_fn(x)
+            if clip_denoised:
+                return x.clamp(-1, 1)
+            return x
+
+        if self.model_mean_type == ModelMeanType.PREVIOUS_X:
+            pred_xstart = process_xstart(
+                self._predict_xstart_from_xprev(x_t=x, t=t, xprev=model_output)
+            )
+            model_mean = model_output
+        elif self.model_mean_type in [ModelMeanType.START_X, ModelMeanType.EPSILON]:
+            if self.model_mean_type == ModelMeanType.START_X:
+                pred_xstart = process_xstart(model_output)
+            else:
+                pred_xstart = process_xstart(
+                    self._predict_xstart_from_eps(x_t=x, t=t, eps=model_output)
+                )
+            model_mean, _, _ = self.q_posterior_mean_variance(x_start=pred_xstart, x_t=x, t=t)
+        else:
+            raise NotImplementedError(self.model_mean_type)
+
+        assert model_mean.shape == model_log_variance.shape == pred_xstart.shape == x.shape
+        return {
+            "mean": model_mean,
+            "variance": model_variance,
+            "log_variance": model_log_variance,
+            "pred_xstart": pred_xstart,
+            "model_output": model_output,
+            "middle_out": middle_out
+        }
+
+    def _predict_xstart_from_eps(self, x_t, t, eps):
+        assert x_t.shape == eps.shape
+        return (
+            _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t
+            - _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape) * eps
+        )
+
+    def _predict_xstart_from_xprev(self, x_t, t, xprev):
+        assert x_t.shape == xprev.shape
+        return (  # (xprev - coef2*x_t) / coef1
+            _extract_into_tensor(1.0 / self.posterior_mean_coef1, t, x_t.shape) * xprev
+            - _extract_into_tensor(
+                self.posterior_mean_coef2 / self.posterior_mean_coef1, t, x_t.shape
+            )
+            * x_t
+        )
+
+    def _predict_eps_from_xstart(self, x_t, t, pred_xstart):
+        return (
+            _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x_t.shape) * x_t - pred_xstart
+        ) / _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x_t.shape)
+
+    def _scale_timesteps(self, t):
+        if self.rescale_timesteps:
+            return t.float() * (1000.0 / self.num_timesteps)
+        return t
+
+    def condition_mean(self, cond_fn, p_mean_var, x, t, model_kwargs=None):
+        """
+        Compute the mean for the previous step, given a function cond_fn that
+        computes the gradient of a conditional log probability with respect to
+        x. In particular, cond_fn computes grad(log(p(y|x))), and we want to
+        condition on y.
+
+        This uses the conditioning strategy from Sohl-Dickstein et al. (2015).
+        """
+        model_kwargs['eps'] = p_mean_var['model_output']
+        #model_kwargs['variance'] = p_mean_var["variance"]
+        gradient = cond_fn(x, self._scale_timesteps(t), **model_kwargs)
+        del model_kwargs['eps']#, model_kwargs['variance']
+        #gradient /= gradient.view(x.shape[0], -1).norm(p=2, dim=1).view(x.shape[0], 1, 1, 1)
+        new_mean = p_mean_var["mean"].float() + p_mean_var["variance"] * gradient.float() #* 200
+        del gradient
+        return new_mean, p_mean_var["variance"]
+
+    def condition_score(self, cond_fn, p_mean_var, x, t, model_kwargs=None):
+        """
+        Compute what the p_mean_variance output would have been, should the
+        model's score function be conditioned by cond_fn.
+
+        See condition_mean() for details on cond_fn.
+
+        Unlike condition_mean(), this instead uses the conditioning strategy
+        from Song et al (2020).
+        """
+        alpha_bar = _extract_into_tensor(self.alphas_cumprod, t, x.shape)
+
+
+        eps = self._predict_eps_from_xstart(x, t, p_mean_var["pred_xstart"])
+        model_kwargs['eps'] = p_mean_var['model_output']
+        gradient = cond_fn(x, self._scale_timesteps(t), **model_kwargs)
+        del model_kwargs['eps']
+
+        eps = eps - (1 - alpha_bar).sqrt() * gradient.float()
+        del gradient
+        out = p_mean_var.copy()
+        out["pred_xstart"] = self._predict_xstart_from_eps(x, t, eps)
+        out["mean"], _, _ = self.q_posterior_mean_variance(x_start=out["pred_xstart"], x_t=x, t=t)
+
+        return out, (1 - alpha_bar).sqrt()
+
+    def p_sample(
+        self, model, x, t, clip_denoised=True, denoised_fn=None, cond_fn=None, model_kwargs=None,
+    ):
+        """
+        Sample x_{t-1} from the model at the given timestep.
+
+        :param model: the model to sample from.
+        :param x: the current tensor at x_{t-1}.
+        :param t: the value of t, starting at 0 for the first diffusion step.
+        :param clip_denoised: if True, clip the x_start prediction to [-1, 1].
+        :param denoised_fn: if not None, a function which applies to the
+            x_start prediction before it is used to sample.
+        :param cond_fn: if not None, this is a gradient function that acts
+                        similarly to the model.
+        :param model_kwargs: if not None, a dict of extra keyword arguments to
+            pass to the model. This can be used for conditioning.
+        :return: a dict containing the following keys:
+                 - 'sample': a random sample from the model.
+                 - 'pred_xstart': a prediction of x_0.
+        """
+        out = self.p_mean_variance(
+            model,
+            x,
+            t,
+            clip_denoised=clip_denoised,
+            denoised_fn=denoised_fn,
+            model_kwargs=model_kwargs,
+        )
+        noise = th.randn_like(x)
+        nonzero_mask = (
+            (t != 0).float().view(-1, *([1] * (len(x.shape) - 1)))
+        )  # no noise when t == 0
+        if cond_fn is not None:
+            out["mean"], factor_ = self.condition_mean(cond_fn, out, x, t, model_kwargs=model_kwargs)
+        sample = out["mean"] + nonzero_mask * th.exp(0.5 * out["log_variance"]) * noise
+        # return {"sample": sample, "pred_xstart": out["pred_xstart"], "factor_": factor_}
+        return {"sample": sample, "pred_xstart": out["pred_xstart"]}
+
+    def p_sample_loop(
+        self,
+        model,
+        shape,
+        noise=None,
+        clip_denoised=True,
+        denoised_fn=None,
+        cond_fn=None,
+        model_kwargs=None,
+        device=None,
+        progress=False,
+        skip_timesteps=0,
+        init_image=None,
+        randomize_class=False,
+        resizers=None,
+        range_t=0,
+
+    ):
+        """
+        Generate samples from the model.
+
+        :param model: the model module.
+        :param shape: the shape of the samples, (N, C, H, W).
+        :param noise: if specified, the noise from the encoder to sample.
+                      Should be of the same shape as `shape`.
+        :param clip_denoised: if True, clip x_start predictions to [-1, 1].
+        :param denoised_fn: if not None, a function which applies to the
+            x_start prediction before it is used to sample.
+        :param cond_fn: if not None, this is a gradient function that acts
+                        similarly to the model.
+        :param model_kwargs: if not None, a dict of extra keyword arguments to
+            pass to the model. This can be used for conditioning.
+        :param device: if specified, the device to create the samples on.
+                       If not specified, use a model parameter's device.
+        :param progress: if True, show a tqdm progress bar.
+        :return: a non-differentiable batch of samples.
+        """
+        final = None
+        for sample in self.p_sample_loop_progressive(
+            model,
+            shape,
+            noise=noise,
+            clip_denoised=clip_denoised,
+            denoised_fn=denoised_fn,
+            cond_fn=cond_fn,
+            model_kwargs=model_kwargs,
+            device=device,
+            progress=progress,
+            skip_timesteps=skip_timesteps,
+            init_image=init_image,
+            randomize_class=randomize_class,
+            resizers=resizers,
+            range_t=range_t
+        ):
+            final = sample
+        return final["sample"]
+
+    def p_sample_loop_progressive(
+        self,
+        model,
+        shape,
+        noise=None,
+        clip_denoised=True,
+        denoised_fn=None,
+        cond_fn=None,
+        model_kwargs=None,
+        device=None,
+        progress=True,
+        skip_timesteps=0,
+        init_image=None,
+        postprocess_fn=None,
+        randomize_class=False,
+        resizers=None,
+        range_t=0,
+        eps_project=30,
+        ilvr_multi=0,
+        seed=1
+    ):
+
+        th.manual_seed(seed)
+        random.seed(seed)
+        np.random.seed(seed)
+        """
+        Generate samples from the model and yield intermediate samples from
+        each timestep of diffusion.
+
+        Arguments are the same as p_sample_loop().
+        Returns a generator over dicts, where each dict is the return value of
+        p_sample().
+        """
+
+        if resizers is not None:
+            down, up = resizers
+
+        if device is None:
+            device = next(model.parameters()).device
+        assert isinstance(shape, (tuple, list))
+        if noise is not None:
+            img = noise
+        else:
+            img = th.randn(*shape, device=device)
+
+        if skip_timesteps and init_image is None:
+            init_image = th.zeros_like(img)
+        # indices = list(range(self.num_timesteps - skip_timesteps))[::-1]
+        indices = [120]
+
+        # scales = [2 ** i for i in range(6)]
+        # scales = np.repeat(scales, len(indices) // len(scales))
+        # scales = list(scales) + [scales[-1]] * (len(indices) - len(scales))
+
+        batch_size = shape[0]
+        #init_image_batch = th.tile(init_image, dims=(batch_size, 1, 1, 1))
+        img = self.q_sample(
+            x_start=init_image, #_batch,
+            t=th.tensor(indices[0], dtype=th.long, device=device),
+            noise=img,
+        )
+        
+
+        #img = init_image + project_perturbation(img - init_image,
+        #                                                       eps=eps_project,
+        #                                                       p=2)  # *out["sample"].abs().max(), p=2)
+
+        if progress:
+            # Lazy import so that we don't depend on tqdm.
+            from tqdm.auto import tqdm
+
+            indices = tqdm(indices)
+
+        if randomize_class and "y" in model_kwargs:
+            print(model.num_classes, model_kwargs["y"].shape, model_kwargs["y"].device)
+            model_kwargs["y"] = th.randint(
+                low=0,
+                high=model.num_classes,
+                size=model_kwargs["y"].shape,
+                device=model_kwargs["y"].device,
+            )
+            print('classes start are', model_kwargs["y"])
+        for i in indices:
+            t = th.tensor([i] * shape[0], device=device)
+            
+            ############start
+            img = self.q_sample(
+                x_start=init_image,
+                t=th.tensor(i, dtype=th.long, device=device),
+                noise=noise
+            )
+            ############end
+            
+            
+            # with th.no_grad():
+            with th.enable_grad():
+                out = self.p_sample(
+                    model,
+                    img,
+                    t,
+                    clip_denoised=clip_denoised,
+                    denoised_fn=denoised_fn,
+                    cond_fn=cond_fn,
+                    model_kwargs=model_kwargs,
+                )
+
+                #### ILVR ####
+                if resizers is not None and False:
+                    if i > range_t:
+                        print('using ILVR, changing scales', i, scales[i], ilvr_multi, out["factor_"].min())
+                        out["sample"] = out["sample"] + ilvr_multi * out["factor_"] * (
+                                    -1 * up(scales[i])(down(scales[i])(out["sample"])) + up(scales[i])(
+                                down(scales[i])(self.q_sample(init_image, t,
+                                                              th.randn(*shape, device=device)))))
+
+
+                if postprocess_fn is not None:
+                    out = postprocess_fn(out, t)
+                yield out, i
+                
+                
+    def tweedie(
+        self,
+        model,
+        shape,
+        model_kwargs=None,
+        device=None,
+        init_image=None,
+        seed=1,
+        indices=[120],
+        noise=None,
+    ):
+        th.manual_seed(seed)
+        random.seed(seed)
+        np.random.seed(seed)
+        batch_size = shape[0]
+        for i in indices:
+            t = th.tensor([i] * shape[0], device=device)
+            
+            # p(xt|x0)
+            if noise is None:
+                noise = th.randn(*shape, device=device)
+            xt = self.q_sample(
+                x_start=init_image,
+                t=th.tensor(i, dtype=th.long, device=device),
+                noise=noise
+            )
+            
+            # tweedie
+            # with th.no_grad():
+            with th.enable_grad():
+                out = self.p_mean_variance(
+                    model,
+                    xt,
+                    t,
+                    model_kwargs=model_kwargs,
+                )
+            # Save the added noise for later on
+            out["noise"] = noise
+            yield out, i
+    
+    def add_noise(
+        self,
+        x0,
+        t,
+    ):
+        ats = self.alphas_cumprod
+        at = th.tensor(ats[t], dtype=th.float32)
+        noise = th.randn_like(x0)
+        xt = at.sqrt() * x0 + (1-at).sqrt() * noise
+        return xt
+    
+    def tweedie_only(
+        self,
+        model,
+        xt,
+        t,
+    ):
+        ats = self.alphas_cumprod
+        at = th.tensor(ats[t], dtype=th.float32)
+        t_in = th.tensor(t, dtype=th.long).unsqueeze(dim=0).to(xt.device)
+        et = model(xt, t_in)
+        if et.shape[1] == 6:
+            et = et[:, :3]
+        x0t = (xt - et * (1-at).sqrt()) / at.sqrt()
+        return x0t
+            
+    def tweedie_simple(
+        self,
+        model,
+        x0,
+        t,
+    ):
+        ats = self.alphas_cumprod
+        at = th.tensor(ats[t], dtype=th.float32)
+        # xt ~ q(xt|x0)
+        # torch.manual_seed(42)
+        noise = th.randn_like(x0)
+        
+        xt = at.sqrt() * x0 + (1 - at).sqrt() * noise
+        
+        # x0 = E(x0|xt)d
+        t_in = th.tensor(t, dtype=th.long).unsqueeze(dim=0).to(x0.device)
+        et = model(xt, t_in)
+        if et.shape[1] == 6:
+            et = et[:, :3]
+        x0t = (xt - et * (1 - at).sqrt()) / at.sqrt()
+        return x0t, xt
+    
+    
+    # @th.no_grad()
+    @th.enable_grad()
+    def sdedit(self, x0, t, model, eta=0.0, skip=5):
+        b = self.betas
+        n = x0.size(0)
+        start_t = (th.ones(n) * t).to("cuda")
+        at_start = compute_alpha(torch.Tensor(b), start_t.long()).to("cuda")
+        xt = at_start.sqrt() * x0 + (1 - at_start).sqrt() * th.randn_like(x0)
+
+        times = list(reversed(np.linspace(0, t, (t//skip)+1)))
+        times = list(map(int, times))
+        
+        for i, t_np in tqdm(enumerate(times[:-1]), total=len(times)):
+            t = (th.ones(n) * times[i]).to("cuda")
+            next_t = (th.ones(n) * times[i+1]).to("cuda")
+            at = compute_alpha(b, t.long()).to("cuda")
+            at_next = compute_alpha(b, next_t.long()).to("cuda")
+
+            et = model(xt, t)
+            et = et[:, :3]
+
+            # 0. x0hat prediction
+            x0_t = (xt - et * (1 - at).sqrt()) / at.sqrt()
+            x0_t = torch.clip(x0_t, -1, 1)
+            # plt.imsave(f"x0t_{i:04d}.png", clear_color(x0_t))
+
+            c1 = ((1 - at / at_next) * (1 - at_next) / (1 - at)).sqrt() * eta
+            c2 = ((1 - at_next) - c1 ** 2).sqrt()
+
+            xt = at_next.sqrt() * x0_t + c1 * th.randn_like(x0_t) + c2 * et
+            # plt.imsave(f"xt_{i:04d}.png", clear_color(xt))
+
+        return xt, x0_t
+
+
+
+    def ddim_sample(
+        self,
+        model,
+        x,
+        t,
+        clip_denoised=True,
+        denoised_fn=None,
+        cond_fn=None,
+        model_kwargs=None,
+        eta=0.0,
+    ):
+        """
+        Sample x_{t-1} from the model using DDIM.
+
+        Same usage as p_sample().
+        """
+        out = self.p_mean_variance(
+            model,
+            x,
+            t,
+            clip_denoised=clip_denoised,
+            denoised_fn=denoised_fn,
+            model_kwargs=model_kwargs,
+        )
+        if cond_fn is not None:
+            out, factor_ = self.condition_score(cond_fn, out, x, t, model_kwargs=model_kwargs)
+
+        # Usually our model outputs epsilon, but we re-derive it
+        # in case we used x_start or x_prev prediction.
+        eps = self._predict_eps_from_xstart(x, t, out["pred_xstart"])
+
+        alpha_bar = _extract_into_tensor(self.alphas_cumprod, t, x.shape)
+        alpha_bar_prev = _extract_into_tensor(self.alphas_cumprod_prev, t, x.shape)
+        sigma = (
+            eta
+            * th.sqrt((1 - alpha_bar_prev) / (1 - alpha_bar))
+            * th.sqrt(1 - alpha_bar / alpha_bar_prev)
+        )
+        # Equation 12.
+        noise = th.randn_like(x)
+        mean_pred = (
+            out["pred_xstart"] * th.sqrt(alpha_bar_prev)
+            + th.sqrt(1 - alpha_bar_prev - sigma ** 2) * eps
+        )
+        nonzero_mask = (
+            (t != 0).float().view(-1, *([1] * (len(x.shape) - 1)))
+        )  # no noise when t == 0
+        sample = mean_pred + nonzero_mask * sigma * noise
+        return {"sample": sample, "pred_xstart": out["pred_xstart"], "factor_": factor_}
+
+    def ddim_reverse_sample(
+        self, model, x, t, clip_denoised=True, denoised_fn=None, model_kwargs=None, eta=0.0,
+    ):
+        """
+        Sample x_{t+1} from the model using DDIM reverse ODE.
+        """
+        assert eta == 0.0, "Reverse ODE only for deterministic path"
+        out = self.p_mean_variance(
+            model,
+            x,
+            t,
+            clip_denoised=clip_denoised,
+            denoised_fn=denoised_fn,
+            model_kwargs=model_kwargs,
+        )
+        # Usually our model outputs epsilon, but we re-derive it
+        # in case we used x_start or x_prev prediction.
+        eps = (
+            _extract_into_tensor(self.sqrt_recip_alphas_cumprod, t, x.shape) * x
+            - out["pred_xstart"]
+        ) / _extract_into_tensor(self.sqrt_recipm1_alphas_cumprod, t, x.shape)
+        alpha_bar_next = _extract_into_tensor(self.alphas_cumprod_next, t, x.shape)
+
+        # Equation 12. reversed
+        mean_pred = out["pred_xstart"] * th.sqrt(alpha_bar_next) + th.sqrt(1 - alpha_bar_next) * eps
+
+        return {"sample": mean_pred, "pred_xstart": out["pred_xstart"]}
+
+    def ddim_sample_loop(
+        self,
+        model,
+        shape,
+        noise=None,
+        clip_denoised=True,
+        denoised_fn=None,
+        cond_fn=None,
+        model_kwargs=None,
+        device=None,
+        progress=False,
+        eta=0.0,
+        skip_timesteps=0,
+        init_image=None,
+        randomize_class=False,
+    ):
+        """
+        Generate samples from the model using DDIM.
+
+        Same usage as p_sample_loop().
+        """
+        final = None
+        for sample in self.ddim_sample_loop_progressive(
+            model,
+            shape,
+            noise=noise,
+            clip_denoised=clip_denoised,
+            denoised_fn=denoised_fn,
+            cond_fn=cond_fn,
+            model_kwargs=model_kwargs,
+            device=device,
+            progress=progress,
+            eta=eta,
+            skip_timesteps=skip_timesteps,
+            init_image=init_image,
+            randomize_class=randomize_class,
+        ):
+            final = sample
+        return final["sample"]
+
+    def ddim_sample_loop_progressive(
+        self,
+        model,
+        shape,
+        noise=None,
+        clip_denoised=True,
+        denoised_fn=None,
+        cond_fn=None,
+        model_kwargs=None,
+        device=None,
+        progress=False,
+        eta=0.0,
+        skip_timesteps=0,
+        init_image=None,
+        randomize_class=False,
+            resizers=None,
+            range_t=0,
+            postprocess_fn=None,
+            eps_project=30,
+            ilvr_multi=0
+
+    ):
+        """
+        Use DDIM to sample from the model and yield intermediate samples from
+        each timestep of DDIM.
+
+        Same usage as p_sample_loop_progressive().
+        """
+        if device is None:
+            device = next(model.parameters()).device
+        assert isinstance(shape, (tuple, list))
+        if noise is not None:
+            img = noise
+        else:
+            img = th.randn(*shape, device=device)
+
+        if resizers is not None:
+            down, up = resizers
+        batch_size = shape[0]
+        #indices = list(range(self.num_timesteps))[::-1]
+        indices = list(range(self.num_timesteps - skip_timesteps))[::-1]
+
+        scales = [2**i for i in range(6)]
+        scales = np.repeat(scales, len(indices) // len(scales))
+        scales = list(scales) + [scales[-1]] * (len(indices) - len(scales))
+
+        #init_image_batch = th.tile(init_image, dims=(batch_size, 1, 1, 1))
+        img = self.q_sample(
+            x_start=init_image,
+            t=th.tensor(indices[0], dtype=th.long, device=device),
+            noise=img,
+        )
+
+        #print('projecting first image', eps_project)
+        #img = init_image + project_perturbation(img - init_image, p=2, eps=eps_project)
+        #img = init_image
+        #print('sampling latents', indices)
+        #for i in indices:
+        #    t = th.tensor([i] * shape[0], device=device)
+        #
+        #    img = self.ddim_reverse_sample(
+        #        model=model,
+        #        x=img,
+        #        t=t
+        #    )["sample"].detach()
+
+        #indices = list(range(self.num_timesteps))[::-1]
+        if progress:
+            # Lazy import so that we don't depend on tqdm.
+            from tqdm.auto import tqdm
+
+            indices = tqdm(indices)
+
+        for i in indices:
+            t = th.tensor([i] * shape[0], device=device)
+            if randomize_class and "y" in model_kwargs:
+                model_kwargs["y"] = th.randint(
+                    low=0,
+                    high=model.num_classes,
+                    size=model_kwargs["y"].shape,
+                    device=model_kwargs["y"].device,
+                )
+            # with th.no_grad():
+            with th.enable_grad():
+                out = self.ddim_sample(
+                    model,
+                    img,
+                    t,
+                    clip_denoised=clip_denoised,
+                    denoised_fn=denoised_fn,
+                    cond_fn=cond_fn,
+                    model_kwargs=model_kwargs,
+                    eta=eta,
+                )
+                yield out
+
+                #### ILVR ####
+                if resizers is not None and False:
+                    if i > range_t:
+                        print('using ILVR, changing scales', i, scales[i], ilvr_multi, out["factor_"].min())
+                        out["sample"] = out["sample"] + ilvr_multi*out["factor_"]*(-1 * up(scales[i])(down(scales[i])(out["sample"])) + up(scales[i])(
+                            down(scales[i])(self.q_sample(init_image, t,
+                                               th.randn(*shape, device=device)))))
+
+                if postprocess_fn is not None:
+                    out = postprocess_fn(out, t)
+
+                img = out["sample"]
+
+    def _vb_terms_bpd(self, model, x_start, x_t, t, clip_denoised=True, model_kwargs=None):
+        """
+        Get a term for the variational lower-bound.
+
+        The resulting units are bits (rather than nats, as one might expect).
+        This allows for comparison to other papers.
+
+        :return: a dict with the following keys:
+                 - 'output': a shape [N] tensor of NLLs or KLs.
+                 - 'pred_xstart': the x_0 predictions.
+        """
+        true_mean, _, true_log_variance_clipped = self.q_posterior_mean_variance(
+            x_start=x_start, x_t=x_t, t=t
+        )
+        out = self.p_mean_variance(
+            model, x_t, t, clip_denoised=clip_denoised, model_kwargs=model_kwargs
+        )
+        kl = normal_kl(true_mean, true_log_variance_clipped, out["mean"], out["log_variance"])
+        kl = mean_flat(kl) / np.log(2.0)
+
+        decoder_nll = -discretized_gaussian_log_likelihood(
+            x_start, means=out["mean"], log_scales=0.5 * out["log_variance"]
+        )
+        assert decoder_nll.shape == x_start.shape
+        decoder_nll = mean_flat(decoder_nll) / np.log(2.0)
+
+        # At the first timestep return the decoder NLL,
+        # otherwise return KL(q(x_{t-1}|x_t,x_0) || p(x_{t-1}|x_t))
+        output = th.where((t == 0), decoder_nll, kl)
+        return {"output": output, "pred_xstart": out["pred_xstart"]}
+
+    def training_losses(self, model, x_start, t, model_kwargs=None, noise=None):
+        """
+        Compute training losses for a single timestep.
+
+        :param model: the model to evaluate loss on.
+        :param x_start: the [N x C x ...] tensor of inputs.
+        :param t: a batch of timestep indices.
+        :param model_kwargs: if not None, a dict of extra keyword arguments to
+            pass to the model. This can be used for conditioning.
+        :param noise: if specified, the specific Gaussian noise to try to remove.
+        :return: a dict with the key "loss" containing a tensor of shape [N].
+                 Some mean or variance settings may also have other keys.
+        """
+        if model_kwargs is None:
+            model_kwargs = {}
+        if noise is None:
+            noise = th.randn_like(x_start)
+        x_t = self.q_sample(x_start, t, noise=noise)
+
+        terms = {}
+
+        if self.loss_type == LossType.KL or self.loss_type == LossType.RESCALED_KL:
+            terms["loss"] = self._vb_terms_bpd(
+                model=model,
+                x_start=x_start,
+                x_t=x_t,
+                t=t,
+                clip_denoised=False,
+                model_kwargs=model_kwargs,
+            )["output"]
+            if self.loss_type == LossType.RESCALED_KL:
+                terms["loss"] *= self.num_timesteps
+        elif self.loss_type == LossType.MSE or self.loss_type == LossType.RESCALED_MSE:
+            model_output = model(x_t, self._scale_timesteps(t), **model_kwargs)
+
+            if self.model_var_type in [
+                ModelVarType.LEARNED,
+                ModelVarType.LEARNED_RANGE,
+            ]:
+                B, C = x_t.shape[:2]
+                assert model_output.shape == (B, C * 2, *x_t.shape[2:])
+                model_output, model_var_values = th.split(model_output, C, dim=1)
+                # Learn the variance using the variational bound, but don't let
+                # it affect our mean prediction.
+                frozen_out = th.cat([model_output.detach(), model_var_values], dim=1)
+                terms["vb"] = self._vb_terms_bpd(
+                    model=lambda *args, r=frozen_out: r,
+                    x_start=x_start,
+                    x_t=x_t,
+                    t=t,
+                    clip_denoised=False,
+                )["output"]
+                if self.loss_type == LossType.RESCALED_MSE:
+                    # Divide by 1000 for equivalence with initial implementation.
+                    # Without a factor of 1/1000, the VB term hurts the MSE term.
+                    terms["vb"] *= self.num_timesteps / 1000.0
+
+            target = {
+                ModelMeanType.PREVIOUS_X: self.q_posterior_mean_variance(
+                    x_start=x_start, x_t=x_t, t=t
+                )[0],
+                ModelMeanType.START_X: x_start,
+                ModelMeanType.EPSILON: noise,
+            }[self.model_mean_type]
+            assert model_output.shape == target.shape == x_start.shape
+            terms["mse"] = mean_flat((target - model_output) ** 2)
+            if "vb" in terms:
+                terms["loss"] = terms["mse"] + terms["vb"]
+            else:
+                terms["loss"] = terms["mse"]
+        else:
+            raise NotImplementedError(self.loss_type)
+
+        return terms
+
+    def _prior_bpd(self, x_start):
+        """
+        Get the prior KL term for the variational lower-bound, measured in
+        bits-per-dim.
+
+        This term can't be optimized, as it only depends on the encoder.
+
+        :param x_start: the [N x C x ...] tensor of inputs.
+        :return: a batch of [N] KL values (in bits), one per batch element.
+        """
+        batch_size = x_start.shape[0]
+        t = th.tensor([self.num_timesteps - 1] * batch_size, device=x_start.device)
+        qt_mean, _, qt_log_variance = self.q_mean_variance(x_start, t)
+        kl_prior = normal_kl(mean1=qt_mean, logvar1=qt_log_variance, mean2=0.0, logvar2=0.0)
+        return mean_flat(kl_prior) / np.log(2.0)
+
+    def calc_bpd_loop(self, model, x_start, clip_denoised=True, model_kwargs=None):
+        """
+        Compute the entire variational lower-bound, measured in bits-per-dim,
+        as well as other related quantities.
+
+        :param model: the model to evaluate loss on.
+        :param x_start: the [N x C x ...] tensor of inputs.
+        :param clip_denoised: if True, clip denoised samples.
+        :param model_kwargs: if not None, a dict of extra keyword arguments to
+            pass to the model. This can be used for conditioning.
+
+        :return: a dict containing the following keys:
+                 - total_bpd: the total variational lower-bound, per batch element.
+                 - prior_bpd: the prior term in the lower-bound.
+                 - vb: an [N x T] tensor of terms in the lower-bound.
+                 - xstart_mse: an [N x T] tensor of x_0 MSEs for each timestep.
+                 - mse: an [N x T] tensor of epsilon MSEs for each timestep.
+        """
+        device = x_start.device
+        batch_size = x_start.shape[0]
+
+        vb = []
+        xstart_mse = []
+        mse = []
+        for t in list(range(self.num_timesteps))[::-1]:
+            t_batch = th.tensor([t] * batch_size, device=device)
+            noise = th.randn_like(x_start)
+            x_t = self.q_sample(x_start=x_start, t=t_batch, noise=noise)
+            # Calculate VLB term at the current timestep
+            # with th.no_grad():
+            with th.enable_grad():
+                out = self._vb_terms_bpd(
+                    model,
+                    x_start=x_start,
+                    x_t=x_t,
+                    t=t_batch,
+                    clip_denoised=clip_denoised,
+                    model_kwargs=model_kwargs,
+                )
+            vb.append(out["output"])
+            xstart_mse.append(mean_flat((out["pred_xstart"] - x_start) ** 2))
+            eps = self._predict_eps_from_xstart(x_t, t_batch, out["pred_xstart"])
+            mse.append(mean_flat((eps - noise) ** 2))
+
+        vb = th.stack(vb, dim=1)
+        xstart_mse = th.stack(xstart_mse, dim=1)
+        mse = th.stack(mse, dim=1)
+
+        prior_bpd = self._prior_bpd(x_start)
+        total_bpd = vb.sum(dim=1) + prior_bpd
+        return {
+            "total_bpd": total_bpd,
+            "prior_bpd": prior_bpd,
+            "vb": vb,
+            "xstart_mse": xstart_mse,
+            "mse": mse,
+        }
+
+
+def _extract_into_tensor(arr, timesteps, broadcast_shape):
+    """
+    Extract values from a 1-D numpy array for a batch of indices.
+
+    :param arr: the 1-D numpy array.
+    :param timesteps: a tensor of indices into the array to extract.
+    :param broadcast_shape: a larger shape of K dimensions with the batch
+                            dimension equal to the length of timesteps.
+    :return: a tensor of shape [batch_size, 1, ...] where the shape has K dims.
+    """
+    res = th.from_numpy(arr).to(device=timesteps.device)[timesteps].float()
+    while len(res.shape) < len(broadcast_shape):
+        res = res[..., None]
+    return res.expand(broadcast_shape)

guided_diffusion/guided_diffusion/image_datasets.py

@@ -0,0 +1,167 @@
+import math
+import random
+
+from PIL import Image
+import blobfile as bf
+from mpi4py import MPI
+import numpy as np
+from torch.utils.data import DataLoader, Dataset
+
+
+def load_data(
+    *,
+    data_dir,
+    batch_size,
+    image_size,
+    class_cond=False,
+    deterministic=False,
+    random_crop=False,
+    random_flip=True,
+):
+    """
+    For a dataset, create a generator over (images, kwargs) pairs.
+
+    Each images is an NCHW float tensor, and the kwargs dict contains zero or
+    more keys, each of which map to a batched Tensor of their own.
+    The kwargs dict can be used for class labels, in which case the key is "y"
+    and the values are integer tensors of class labels.
+
+    :param data_dir: a dataset directory.
+    :param batch_size: the batch size of each returned pair.
+    :param image_size: the size to which images are resized.
+    :param class_cond: if True, include a "y" key in returned dicts for class
+                       label. If classes are not available and this is true, an
+                       exception will be raised.
+    :param deterministic: if True, yield results in a deterministic order.
+    :param random_crop: if True, randomly crop the images for augmentation.
+    :param random_flip: if True, randomly flip the images for augmentation.
+    """
+    if not data_dir:
+        raise ValueError("unspecified data directory")
+    all_files = _list_image_files_recursively(data_dir)
+    classes = None
+    if class_cond:
+        # Assume classes are the first part of the filename,
+        # before an underscore.
+        class_names = [bf.basename(path).split("_")[0] for path in all_files]
+        sorted_classes = {x: i for i, x in enumerate(sorted(set(class_names)))}
+        classes = [sorted_classes[x] for x in class_names]
+    dataset = ImageDataset(
+        image_size,
+        all_files,
+        classes=classes,
+        shard=MPI.COMM_WORLD.Get_rank(),
+        num_shards=MPI.COMM_WORLD.Get_size(),
+        random_crop=random_crop,
+        random_flip=random_flip,
+    )
+    if deterministic:
+        loader = DataLoader(
+            dataset, batch_size=batch_size, shuffle=False, num_workers=1, drop_last=True
+        )
+    else:
+        loader = DataLoader(
+            dataset, batch_size=batch_size, shuffle=True, num_workers=1, drop_last=True
+        )
+    while True:
+        yield from loader
+
+
+def _list_image_files_recursively(data_dir):
+    results = []
+    for entry in sorted(bf.listdir(data_dir)):
+        full_path = bf.join(data_dir, entry)
+        ext = entry.split(".")[-1]
+        if "." in entry and ext.lower() in ["jpg", "jpeg", "png", "gif"]:
+            results.append(full_path)
+        elif bf.isdir(full_path):
+            results.extend(_list_image_files_recursively(full_path))
+    return results
+
+
+class ImageDataset(Dataset):
+    def __init__(
+        self,
+        resolution,
+        image_paths,
+        classes=None,
+        shard=0,
+        num_shards=1,
+        random_crop=False,
+        random_flip=True,
+    ):
+        super().__init__()
+        self.resolution = resolution
+        self.local_images = image_paths[shard:][::num_shards]
+        self.local_classes = None if classes is None else classes[shard:][::num_shards]
+        self.random_crop = random_crop
+        self.random_flip = random_flip
+
+    def __len__(self):
+        return len(self.local_images)
+
+    def __getitem__(self, idx):
+        path = self.local_images[idx]
+        with bf.BlobFile(path, "rb") as f:
+            pil_image = Image.open(f)
+            pil_image.load()
+        pil_image = pil_image.convert("RGB")
+
+        if self.random_crop:
+            arr = random_crop_arr(pil_image, self.resolution)
+        else:
+            arr = center_crop_arr(pil_image, self.resolution)
+
+        if self.random_flip and random.random() < 0.5:
+            arr = arr[:, ::-1]
+
+        arr = arr.astype(np.float32) / 127.5 - 1
+
+        out_dict = {}
+        if self.local_classes is not None:
+            out_dict["y"] = np.array(self.local_classes[idx], dtype=np.int64)
+        return np.transpose(arr, [2, 0, 1]), out_dict
+
+
+def center_crop_arr(pil_image, image_size):
+    # We are not on a new enough PIL to support the `reducing_gap`
+    # argument, which uses BOX downsampling at powers of two first.
+    # Thus, we do it by hand to improve downsample quality.
+    while min(*pil_image.size) >= 2 * image_size:
+        pil_image = pil_image.resize(
+            tuple(x // 2 for x in pil_image.size), resample=Image.BOX
+        )
+
+    scale = image_size / min(*pil_image.size)
+    pil_image = pil_image.resize(
+        tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC
+    )
+
+    arr = np.array(pil_image)
+    crop_y = (arr.shape[0] - image_size) // 2
+    crop_x = (arr.shape[1] - image_size) // 2
+    return arr[crop_y : crop_y + image_size, crop_x : crop_x + image_size]
+
+
+def random_crop_arr(pil_image, image_size, min_crop_frac=0.8, max_crop_frac=1.0):
+    min_smaller_dim_size = math.ceil(image_size / max_crop_frac)
+    max_smaller_dim_size = math.ceil(image_size / min_crop_frac)
+    smaller_dim_size = random.randrange(min_smaller_dim_size, max_smaller_dim_size + 1)
+
+    # We are not on a new enough PIL to support the `reducing_gap`
+    # argument, which uses BOX downsampling at powers of two first.
+    # Thus, we do it by hand to improve downsample quality.
+    while min(*pil_image.size) >= 2 * smaller_dim_size:
+        pil_image = pil_image.resize(
+            tuple(x // 2 for x in pil_image.size), resample=Image.BOX
+        )
+
+    scale = smaller_dim_size / min(*pil_image.size)
+    pil_image = pil_image.resize(
+        tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC
+    )
+
+    arr = np.array(pil_image)
+    crop_y = random.randrange(arr.shape[0] - image_size + 1)
+    crop_x = random.randrange(arr.shape[1] - image_size + 1)
+    return arr[crop_y : crop_y + image_size, crop_x : crop_x + image_size]

guided_diffusion/guided_diffusion/logger.py

@@ -0,0 +1,495 @@
+"""
+Logger copied from OpenAI baselines to avoid extra RL-based dependencies:
+https://github.com/openai/baselines/blob/ea25b9e8b234e6ee1bca43083f8f3cf974143998/baselines/logger.py
+"""
+
+import os
+import sys
+import shutil
+import os.path as osp
+import json
+import time
+import datetime
+import tempfile
+import warnings
+from collections import defaultdict
+from contextlib import contextmanager
+
+DEBUG = 10
+INFO = 20
+WARN = 30
+ERROR = 40
+
+DISABLED = 50
+
+
+class KVWriter(object):
+    def writekvs(self, kvs):
+        raise NotImplementedError
+
+
+class SeqWriter(object):
+    def writeseq(self, seq):
+        raise NotImplementedError
+
+
+class HumanOutputFormat(KVWriter, SeqWriter):
+    def __init__(self, filename_or_file):
+        if isinstance(filename_or_file, str):
+            self.file = open(filename_or_file, "wt")
+            self.own_file = True
+        else:
+            assert hasattr(filename_or_file, "read"), (
+                "expected file or str, got %s" % filename_or_file
+            )
+            self.file = filename_or_file
+            self.own_file = False
+
+    def writekvs(self, kvs):
+        # Create strings for printing
+        key2str = {}
+        for (key, val) in sorted(kvs.items()):
+            if hasattr(val, "__float__"):
+                valstr = "%-8.3g" % val
+            else:
+                valstr = str(val)
+            key2str[self._truncate(key)] = self._truncate(valstr)
+
+        # Find max widths
+        if len(key2str) == 0:
+            print("WARNING: tried to write empty key-value dict")
+            return
+        else:
+            keywidth = max(map(len, key2str.keys()))
+            valwidth = max(map(len, key2str.values()))
+
+        # Write out the data
+        dashes = "-" * (keywidth + valwidth + 7)
+        lines = [dashes]
+        for (key, val) in sorted(key2str.items(), key=lambda kv: kv[0].lower()):
+            lines.append(
+                "| %s%s | %s%s |"
+                % (key, " " * (keywidth - len(key)), val, " " * (valwidth - len(val)))
+            )
+        lines.append(dashes)
+        self.file.write("\n".join(lines) + "\n")
+
+        # Flush the output to the file
+        self.file.flush()
+
+    def _truncate(self, s):
+        maxlen = 30
+        return s[: maxlen - 3] + "..." if len(s) > maxlen else s
+
+    def writeseq(self, seq):
+        seq = list(seq)
+        for (i, elem) in enumerate(seq):
+            self.file.write(elem)
+            if i < len(seq) - 1:  # add space unless this is the last one
+                self.file.write(" ")
+        self.file.write("\n")
+        self.file.flush()
+
+    def close(self):
+        if self.own_file:
+            self.file.close()
+
+
+class JSONOutputFormat(KVWriter):
+    def __init__(self, filename):
+        self.file = open(filename, "wt")
+
+    def writekvs(self, kvs):
+        for k, v in sorted(kvs.items()):
+            if hasattr(v, "dtype"):
+                kvs[k] = float(v)
+        self.file.write(json.dumps(kvs) + "\n")
+        self.file.flush()
+
+    def close(self):
+        self.file.close()
+
+
+class CSVOutputFormat(KVWriter):
+    def __init__(self, filename):
+        self.file = open(filename, "w+t")
+        self.keys = []
+        self.sep = ","
+
+    def writekvs(self, kvs):
+        # Add our current row to the history
+        extra_keys = list(kvs.keys() - self.keys)
+        extra_keys.sort()
+        if extra_keys:
+            self.keys.extend(extra_keys)
+            self.file.seek(0)
+            lines = self.file.readlines()
+            self.file.seek(0)
+            for (i, k) in enumerate(self.keys):
+                if i > 0:
+                    self.file.write(",")
+                self.file.write(k)
+            self.file.write("\n")
+            for line in lines[1:]:
+                self.file.write(line[:-1])
+                self.file.write(self.sep * len(extra_keys))
+                self.file.write("\n")
+        for (i, k) in enumerate(self.keys):
+            if i > 0:
+                self.file.write(",")
+            v = kvs.get(k)
+            if v is not None:
+                self.file.write(str(v))
+        self.file.write("\n")
+        self.file.flush()
+
+    def close(self):
+        self.file.close()
+
+
+class TensorBoardOutputFormat(KVWriter):
+    """
+    Dumps key/value pairs into TensorBoard's numeric format.
+    """
+
+    def __init__(self, dir):
+        os.makedirs(dir, exist_ok=True)
+        self.dir = dir
+        self.step = 1
+        prefix = "events"
+        path = osp.join(osp.abspath(dir), prefix)
+        import tensorflow as tf
+        from tensorflow.python import pywrap_tensorflow
+        from tensorflow.core.util import event_pb2
+        from tensorflow.python.util import compat
+
+        self.tf = tf
+        self.event_pb2 = event_pb2
+        self.pywrap_tensorflow = pywrap_tensorflow
+        self.writer = pywrap_tensorflow.EventsWriter(compat.as_bytes(path))
+
+    def writekvs(self, kvs):
+        def summary_val(k, v):
+            kwargs = {"tag": k, "simple_value": float(v)}
+            return self.tf.Summary.Value(**kwargs)
+
+        summary = self.tf.Summary(value=[summary_val(k, v) for k, v in kvs.items()])
+        event = self.event_pb2.Event(wall_time=time.time(), summary=summary)
+        event.step = (
+            self.step
+        )  # is there any reason why you'd want to specify the step?
+        self.writer.WriteEvent(event)
+        self.writer.Flush()
+        self.step += 1
+
+    def close(self):
+        if self.writer:
+            self.writer.Close()
+            self.writer = None
+
+
+def make_output_format(format, ev_dir, log_suffix=""):
+    os.makedirs(ev_dir, exist_ok=True)
+    if format == "stdout":
+        return HumanOutputFormat(sys.stdout)
+    elif format == "log":
+        return HumanOutputFormat(osp.join(ev_dir, "log%s.txt" % log_suffix))
+    elif format == "json":
+        return JSONOutputFormat(osp.join(ev_dir, "progress%s.json" % log_suffix))
+    elif format == "csv":
+        return CSVOutputFormat(osp.join(ev_dir, "progress%s.csv" % log_suffix))
+    elif format == "tensorboard":
+        return TensorBoardOutputFormat(osp.join(ev_dir, "tb%s" % log_suffix))
+    else:
+        raise ValueError("Unknown format specified: %s" % (format,))
+
+
+# ================================================================
+# API
+# ================================================================
+
+
+def logkv(key, val):
+    """
+    Log a value of some diagnostic
+    Call this once for each diagnostic quantity, each iteration
+    If called many times, last value will be used.
+    """
+    get_current().logkv(key, val)
+
+
+def logkv_mean(key, val):
+    """
+    The same as logkv(), but if called many times, values averaged.
+    """
+    get_current().logkv_mean(key, val)
+
+
+def logkvs(d):
+    """
+    Log a dictionary of key-value pairs
+    """
+    for (k, v) in d.items():
+        logkv(k, v)
+
+
+def dumpkvs():
+    """
+    Write all of the diagnostics from the current iteration
+    """
+    return get_current().dumpkvs()
+
+
+def getkvs():
+    return get_current().name2val
+
+
+def log(*args, level=INFO):
+    """
+    Write the sequence of args, with no separators, to the console and output files (if you've configured an output file).
+    """
+    get_current().log(*args, level=level)
+
+
+def debug(*args):
+    log(*args, level=DEBUG)
+
+
+def info(*args):
+    log(*args, level=INFO)
+
+
+def warn(*args):
+    log(*args, level=WARN)
+
+
+def error(*args):
+    log(*args, level=ERROR)
+
+
+def set_level(level):
+    """
+    Set logging threshold on current logger.
+    """
+    get_current().set_level(level)
+
+
+def set_comm(comm):
+    get_current().set_comm(comm)
+
+
+def get_dir():
+    """
+    Get directory that log files are being written to.
+    will be None if there is no output directory (i.e., if you didn't call start)
+    """
+    return get_current().get_dir()
+
+
+record_tabular = logkv
+dump_tabular = dumpkvs
+
+
+@contextmanager
+def profile_kv(scopename):
+    logkey = "wait_" + scopename
+    tstart = time.time()
+    try:
+        yield
+    finally:
+        get_current().name2val[logkey] += time.time() - tstart
+
+
+def profile(n):
+    """
+    Usage:
+    @profile("my_func")
+    def my_func(): code
+    """
+
+    def decorator_with_name(func):
+        def func_wrapper(*args, **kwargs):
+            with profile_kv(n):
+                return func(*args, **kwargs)
+
+        return func_wrapper
+
+    return decorator_with_name
+
+
+# ================================================================
+# Backend
+# ================================================================
+
+
+def get_current():
+    if Logger.CURRENT is None:
+        _configure_default_logger()
+
+    return Logger.CURRENT
+
+
+class Logger(object):
+    DEFAULT = None  # A logger with no output files. (See right below class definition)
+    # So that you can still log to the terminal without setting up any output files
+    CURRENT = None  # Current logger being used by the free functions above
+
+    def __init__(self, dir, output_formats, comm=None):
+        self.name2val = defaultdict(float)  # values this iteration
+        self.name2cnt = defaultdict(int)
+        self.level = INFO
+        self.dir = dir
+        self.output_formats = output_formats
+        self.comm = comm
+
+    # Logging API, forwarded
+    # ----------------------------------------
+    def logkv(self, key, val):
+        self.name2val[key] = val
+
+    def logkv_mean(self, key, val):
+        oldval, cnt = self.name2val[key], self.name2cnt[key]
+        self.name2val[key] = oldval * cnt / (cnt + 1) + val / (cnt + 1)
+        self.name2cnt[key] = cnt + 1
+
+    def dumpkvs(self):
+        if self.comm is None:
+            d = self.name2val
+        else:
+            d = mpi_weighted_mean(
+                self.comm,
+                {
+                    name: (val, self.name2cnt.get(name, 1))
+                    for (name, val) in self.name2val.items()
+                },
+            )
+            if self.comm.rank != 0:
+                d["dummy"] = 1  # so we don't get a warning about empty dict
+        out = d.copy()  # Return the dict for unit testing purposes
+        for fmt in self.output_formats:
+            if isinstance(fmt, KVWriter):
+                fmt.writekvs(d)
+        self.name2val.clear()
+        self.name2cnt.clear()
+        return out
+
+    def log(self, *args, level=INFO):
+        if self.level <= level:
+            self._do_log(args)
+
+    # Configuration
+    # ----------------------------------------
+    def set_level(self, level):
+        self.level = level
+
+    def set_comm(self, comm):
+        self.comm = comm
+
+    def get_dir(self):
+        return self.dir
+
+    def close(self):
+        for fmt in self.output_formats:
+            fmt.close()
+
+    # Misc
+    # ----------------------------------------
+    def _do_log(self, args):
+        for fmt in self.output_formats:
+            if isinstance(fmt, SeqWriter):
+                fmt.writeseq(map(str, args))
+
+
+def get_rank_without_mpi_import():
+    # check environment variables here instead of importing mpi4py
+    # to avoid calling MPI_Init() when this module is imported
+    for varname in ["PMI_RANK", "OMPI_COMM_WORLD_RANK"]:
+        if varname in os.environ:
+            return int(os.environ[varname])
+    return 0
+
+
+def mpi_weighted_mean(comm, local_name2valcount):
+    """
+    Copied from: https://github.com/openai/baselines/blob/ea25b9e8b234e6ee1bca43083f8f3cf974143998/baselines/common/mpi_util.py#L110
+    Perform a weighted average over dicts that are each on a different node
+    Input: local_name2valcount: dict mapping key -> (value, count)
+    Returns: key -> mean
+    """
+    all_name2valcount = comm.gather(local_name2valcount)
+    if comm.rank == 0:
+        name2sum = defaultdict(float)
+        name2count = defaultdict(float)
+        for n2vc in all_name2valcount:
+            for (name, (val, count)) in n2vc.items():
+                try:
+                    val = float(val)
+                except ValueError:
+                    if comm.rank == 0:
+                        warnings.warn(
+                            "WARNING: tried to compute mean on non-float {}={}".format(
+                                name, val
+                            )
+                        )
+                else:
+                    name2sum[name] += val * count
+                    name2count[name] += count
+        return {name: name2sum[name] / name2count[name] for name in name2sum}
+    else:
+        return {}
+
+
+def configure(dir=None, format_strs=None, comm=None, log_suffix=""):
+    """
+    If comm is provided, average all numerical stats across that comm
+    """
+    if dir is None:
+        dir = os.getenv("OPENAI_LOGDIR")
+    if dir is None:
+        dir = osp.join(
+            tempfile.gettempdir(),
+            datetime.datetime.now().strftime("openai-%Y-%m-%d-%H-%M-%S-%f"),
+        )
+    assert isinstance(dir, str)
+    dir = os.path.expanduser(dir)
+    os.makedirs(os.path.expanduser(dir), exist_ok=True)
+
+    rank = get_rank_without_mpi_import()
+    if rank > 0:
+        log_suffix = log_suffix + "-rank%03i" % rank
+
+    if format_strs is None:
+        if rank == 0:
+            format_strs = os.getenv("OPENAI_LOG_FORMAT", "stdout,log,csv").split(",")
+        else:
+            format_strs = os.getenv("OPENAI_LOG_FORMAT_MPI", "log").split(",")
+    format_strs = filter(None, format_strs)
+    output_formats = [make_output_format(f, dir, log_suffix) for f in format_strs]
+
+    Logger.CURRENT = Logger(dir=dir, output_formats=output_formats, comm=comm)
+    if output_formats:
+        log("Logging to %s" % dir)
+
+
+def _configure_default_logger():
+    configure()
+    Logger.DEFAULT = Logger.CURRENT
+
+
+def reset():
+    if Logger.CURRENT is not Logger.DEFAULT:
+        Logger.CURRENT.close()
+        Logger.CURRENT = Logger.DEFAULT
+        log("Reset logger")
+
+
+@contextmanager
+def scoped_configure(dir=None, format_strs=None, comm=None):
+    prevlogger = Logger.CURRENT
+    configure(dir=dir, format_strs=format_strs, comm=comm)
+    try:
+        yield
+    finally:
+        Logger.CURRENT.close()
+        Logger.CURRENT = prevlogger
+

guided_diffusion/guided_diffusion/losses.py

@@ -0,0 +1,77 @@
+"""
+Helpers for various likelihood-based losses. These are ported from the original
+Ho et al. diffusion models codebase:
+https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/utils.py
+"""
+
+import numpy as np
+
+import torch as th
+
+
+def normal_kl(mean1, logvar1, mean2, logvar2):
+    """
+    Compute the KL divergence between two gaussians.
+
+    Shapes are automatically broadcasted, so batches can be compared to
+    scalars, among other use cases.
+    """
+    tensor = None
+    for obj in (mean1, logvar1, mean2, logvar2):
+        if isinstance(obj, th.Tensor):
+            tensor = obj
+            break
+    assert tensor is not None, "at least one argument must be a Tensor"
+
+    # Force variances to be Tensors. Broadcasting helps convert scalars to
+    # Tensors, but it does not work for th.exp().
+    logvar1, logvar2 = [
+        x if isinstance(x, th.Tensor) else th.tensor(x).to(tensor)
+        for x in (logvar1, logvar2)
+    ]
+
+    return 0.5 * (
+        -1.0
+        + logvar2
+        - logvar1
+        + th.exp(logvar1 - logvar2)
+        + ((mean1 - mean2) ** 2) * th.exp(-logvar2)
+    )
+
+
+def approx_standard_normal_cdf(x):
+    """
+    A fast approximation of the cumulative distribution function of the
+    standard normal.
+    """
+    return 0.5 * (1.0 + th.tanh(np.sqrt(2.0 / np.pi) * (x + 0.044715 * th.pow(x, 3))))
+
+
+def discretized_gaussian_log_likelihood(x, *, means, log_scales):
+    """
+    Compute the log-likelihood of a Gaussian distribution discretizing to a
+    given image.
+
+    :param x: the target images. It is assumed that this was uint8 values,
+              rescaled to the range [-1, 1].
+    :param means: the Gaussian mean Tensor.
+    :param log_scales: the Gaussian log stddev Tensor.
+    :return: a tensor like x of log probabilities (in nats).
+    """
+    assert x.shape == means.shape == log_scales.shape
+    centered_x = x - means
+    inv_stdv = th.exp(-log_scales)
+    plus_in = inv_stdv * (centered_x + 1.0 / 255.0)
+    cdf_plus = approx_standard_normal_cdf(plus_in)
+    min_in = inv_stdv * (centered_x - 1.0 / 255.0)
+    cdf_min = approx_standard_normal_cdf(min_in)
+    log_cdf_plus = th.log(cdf_plus.clamp(min=1e-12))
+    log_one_minus_cdf_min = th.log((1.0 - cdf_min).clamp(min=1e-12))
+    cdf_delta = cdf_plus - cdf_min
+    log_probs = th.where(
+        x < -0.999,
+        log_cdf_plus,
+        th.where(x > 0.999, log_one_minus_cdf_min, th.log(cdf_delta.clamp(min=1e-12))),
+    )
+    assert log_probs.shape == x.shape
+    return log_probs

guided_diffusion/guided_diffusion/nn.py

@@ -0,0 +1,170 @@
+"""
+Various utilities for neural networks.
+"""
+
+import math
+
+import torch as th
+import torch.nn as nn
+
+
+# PyTorch 1.7 has SiLU, but we support PyTorch 1.5.
+class SiLU(nn.Module):
+    def forward(self, x):
+        return x * th.sigmoid(x)
+
+
+class GroupNorm32(nn.GroupNorm):
+    def forward(self, x):
+        return super().forward(x.float()).type(x.dtype)
+
+
+def conv_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D convolution module.
+    """
+    if dims == 1:
+        return nn.Conv1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.Conv2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.Conv3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def linear(*args, **kwargs):
+    """
+    Create a linear module.
+    """
+    return nn.Linear(*args, **kwargs)
+
+
+def avg_pool_nd(dims, *args, **kwargs):
+    """
+    Create a 1D, 2D, or 3D average pooling module.
+    """
+    if dims == 1:
+        return nn.AvgPool1d(*args, **kwargs)
+    elif dims == 2:
+        return nn.AvgPool2d(*args, **kwargs)
+    elif dims == 3:
+        return nn.AvgPool3d(*args, **kwargs)
+    raise ValueError(f"unsupported dimensions: {dims}")
+
+
+def update_ema(target_params, source_params, rate=0.99):
+    """
+    Update target parameters to be closer to those of source parameters using
+    an exponential moving average.
+
+    :param target_params: the target parameter sequence.
+    :param source_params: the source parameter sequence.
+    :param rate: the EMA rate (closer to 1 means slower).
+    """
+    for targ, src in zip(target_params, source_params):
+        targ.detach().mul_(rate).add_(src, alpha=1 - rate)
+
+
+def zero_module(module):
+    """
+    Zero out the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().zero_()
+    return module
+
+
+def scale_module(module, scale):
+    """
+    Scale the parameters of a module and return it.
+    """
+    for p in module.parameters():
+        p.detach().mul_(scale)
+    return module
+
+
+def mean_flat(tensor):
+    """
+    Take the mean over all non-batch dimensions.
+    """
+    return tensor.mean(dim=list(range(1, len(tensor.shape))))
+
+
+def normalization(channels):
+    """
+    Make a standard normalization layer.
+
+    :param channels: number of input channels.
+    :return: an nn.Module for normalization.
+    """
+    return GroupNorm32(32, channels)
+
+
+def timestep_embedding(timesteps, dim, max_period=10000):
+    """
+    Create sinusoidal timestep embeddings.
+
+    :param timesteps: a 1-D Tensor of N indices, one per batch element.
+                      These may be fractional.
+    :param dim: the dimension of the output.
+    :param max_period: controls the minimum frequency of the embeddings.
+    :return: an [N x dim] Tensor of positional embeddings.
+    """
+    half = dim // 2
+    freqs = th.exp(
+        -math.log(max_period) * th.arange(start=0, end=half, dtype=th.float32) / half
+    ).to(device=timesteps.device)
+    args = timesteps[:, None].float() * freqs[None]
+    embedding = th.cat([th.cos(args), th.sin(args)], dim=-1)
+    if dim % 2:
+        embedding = th.cat([embedding, th.zeros_like(embedding[:, :1])], dim=-1)
+    return embedding
+
+
+def checkpoint(func, inputs, params, flag):
+    """
+    Evaluate a function without caching intermediate activations, allowing for
+    reduced memory at the expense of extra compute in the backward pass.
+
+    :param func: the function to evaluate.
+    :param inputs: the argument sequence to pass to `func`.
+    :param params: a sequence of parameters `func` depends on but does not
+                   explicitly take as arguments.
+    :param flag: if False, disable gradient checkpointing.
+    """
+    if flag:
+        args = tuple(inputs) + tuple(params)
+        return CheckpointFunction.apply(func, len(inputs), *args)
+    else:
+        return func(*inputs)
+
+
+class CheckpointFunction(th.autograd.Function):
+    @staticmethod
+    def forward(ctx, run_function, length, *args):
+        ctx.run_function = run_function
+        ctx.input_tensors = list(args[:length])
+        ctx.input_params = list(args[length:])
+        with th.no_grad():
+            output_tensors = ctx.run_function(*ctx.input_tensors)
+        return output_tensors
+
+    @staticmethod
+    def backward(ctx, *output_grads):
+        ctx.input_tensors = [x.detach().requires_grad_(True) for x in ctx.input_tensors]
+        with th.enable_grad():
+            # Fixes a bug where the first op in run_function modifies the
+            # Tensor storage in place, which is not allowed for detach()'d
+            # Tensors.
+            shallow_copies = [x.view_as(x) for x in ctx.input_tensors]
+            output_tensors = ctx.run_function(*shallow_copies)
+        input_grads = th.autograd.grad(
+            output_tensors,
+            ctx.input_tensors + ctx.input_params,
+            output_grads,
+            allow_unused=True,
+        )
+        #del ctx.input_tensors
+        #del ctx.input_params
+        del output_tensors
+        return (None, None) + input_grads

guided_diffusion/guided_diffusion/resample.py

@@ -0,0 +1,154 @@
+from abc import ABC, abstractmethod
+
+import numpy as np
+import torch as th
+import torch.distributed as dist
+
+
+def create_named_schedule_sampler(name, diffusion):
+    """
+    Create a ScheduleSampler from a library of pre-defined samplers.
+
+    :param name: the name of the sampler.
+    :param diffusion: the diffusion object to sample for.
+    """
+    if name == "uniform":
+        return UniformSampler(diffusion)
+    elif name == "loss-second-moment":
+        return LossSecondMomentResampler(diffusion)
+    else:
+        raise NotImplementedError(f"unknown schedule sampler: {name}")
+
+
+class ScheduleSampler(ABC):
+    """
+    A distribution over timesteps in the diffusion process, intended to reduce
+    variance of the objective.
+
+    By default, samplers perform unbiased importance sampling, in which the
+    objective's mean is unchanged.
+    However, subclasses may override sample() to change how the resampled
+    terms are reweighted, allowing for actual changes in the objective.
+    """
+
+    @abstractmethod
+    def weights(self):
+        """
+        Get a numpy array of weights, one per diffusion step.
+
+        The weights needn't be normalized, but must be positive.
+        """
+
+    def sample(self, batch_size, device):
+        """
+        Importance-sample timesteps for a batch.
+
+        :param batch_size: the number of timesteps.
+        :param device: the torch device to save to.
+        :return: a tuple (timesteps, weights):
+                 - timesteps: a tensor of timestep indices.
+                 - weights: a tensor of weights to scale the resulting losses.
+        """
+        w = self.weights()
+        p = w / np.sum(w)
+        indices_np = np.random.choice(len(p), size=(batch_size,), p=p)
+        indices = th.from_numpy(indices_np).long().to(device)
+        weights_np = 1 / (len(p) * p[indices_np])
+        weights = th.from_numpy(weights_np).float().to(device)
+        return indices, weights
+
+
+class UniformSampler(ScheduleSampler):
+    def __init__(self, diffusion):
+        self.diffusion = diffusion
+        self._weights = np.ones([diffusion.num_timesteps])
+
+    def weights(self):
+        return self._weights
+
+
+class LossAwareSampler(ScheduleSampler):
+    def update_with_local_losses(self, local_ts, local_losses):
+        """
+        Update the reweighting using losses from a model.
+
+        Call this method from each rank with a batch of timesteps and the
+        corresponding losses for each of those timesteps.
+        This method will perform synchronization to make sure all of the ranks
+        maintain the exact same reweighting.
+
+        :param local_ts: an integer Tensor of timesteps.
+        :param local_losses: a 1D Tensor of losses.
+        """
+        batch_sizes = [
+            th.tensor([0], dtype=th.int32, device=local_ts.device)
+            for _ in range(dist.get_world_size())
+        ]
+        dist.all_gather(
+            batch_sizes,
+            th.tensor([len(local_ts)], dtype=th.int32, device=local_ts.device),
+        )
+
+        # Pad all_gather batches to be the maximum batch size.
+        batch_sizes = [x.item() for x in batch_sizes]
+        max_bs = max(batch_sizes)
+
+        timestep_batches = [th.zeros(max_bs).to(local_ts) for bs in batch_sizes]
+        loss_batches = [th.zeros(max_bs).to(local_losses) for bs in batch_sizes]
+        dist.all_gather(timestep_batches, local_ts)
+        dist.all_gather(loss_batches, local_losses)
+        timesteps = [
+            x.item() for y, bs in zip(timestep_batches, batch_sizes) for x in y[:bs]
+        ]
+        losses = [x.item() for y, bs in zip(loss_batches, batch_sizes) for x in y[:bs]]
+        self.update_with_all_losses(timesteps, losses)
+
+    @abstractmethod
+    def update_with_all_losses(self, ts, losses):
+        """
+        Update the reweighting using losses from a model.
+
+        Sub-classes should override this method to update the reweighting
+        using losses from the model.
+
+        This method directly updates the reweighting without synchronizing
+        between workers. It is called by update_with_local_losses from all
+        ranks with identical arguments. Thus, it should have deterministic
+        behavior to maintain state across workers.
+
+        :param ts: a list of int timesteps.
+        :param losses: a list of float losses, one per timestep.
+        """
+
+
+class LossSecondMomentResampler(LossAwareSampler):
+    def __init__(self, diffusion, history_per_term=10, uniform_prob=0.001):
+        self.diffusion = diffusion
+        self.history_per_term = history_per_term
+        self.uniform_prob = uniform_prob
+        self._loss_history = np.zeros(
+            [diffusion.num_timesteps, history_per_term], dtype=np.float64
+        )
+        self._loss_counts = np.zeros([diffusion.num_timesteps], dtype=np.int)
+
+    def weights(self):
+        if not self._warmed_up():
+            return np.ones([self.diffusion.num_timesteps], dtype=np.float64)
+        weights = np.sqrt(np.mean(self._loss_history ** 2, axis=-1))
+        weights /= np.sum(weights)
+        weights *= 1 - self.uniform_prob
+        weights += self.uniform_prob / len(weights)
+        return weights
+
+    def update_with_all_losses(self, ts, losses):
+        for t, loss in zip(ts, losses):
+            if self._loss_counts[t] == self.history_per_term:
+                # Shift out the oldest loss term.
+                self._loss_history[t, :-1] = self._loss_history[t, 1:]
+                self._loss_history[t, -1] = loss
+            else:
+                self._loss_history[t, self._loss_counts[t]] = loss
+                self._loss_counts[t] += 1
+
+    def _warmed_up(self):
+        return (self._loss_counts == self.history_per_term).all()

guided_diffusion/guided_diffusion/respace.py

@@ -0,0 +1,128 @@
+import numpy as np
+import torch as th
+
+from .gaussian_diffusion import GaussianDiffusion
+
+
+def space_timesteps(num_timesteps, section_counts):
+    """
+    Create a list of timesteps to use from an original diffusion process,
+    given the number of timesteps we want to take from equally-sized portions
+    of the original process.
+
+    For example, if there's 300 timesteps and the section counts are [10,15,20]
+    then the first 100 timesteps are strided to be 10 timesteps, the second 100
+    are strided to be 15 timesteps, and the final 100 are strided to be 20.
+
+    If the stride is a string starting with "ddim", then the fixed striding
+    from the DDIM paper is used, and only one section is allowed.
+
+    :param num_timesteps: the number of diffusion steps in the original
+                          process to divide up.
+    :param section_counts: either a list of numbers, or a string containing
+                           comma-separated numbers, indicating the step count
+                           per section. As a special case, use "ddimN" where N
+                           is a number of steps to use the striding from the
+                           DDIM paper.
+    :return: a set of diffusion steps from the original process to use.
+    """
+    if isinstance(section_counts, str):
+        if section_counts.startswith("ddim"):
+            desired_count = int(section_counts[len("ddim") :])
+            for i in range(1, num_timesteps):
+                if len(range(0, num_timesteps, i)) == desired_count:
+                    return set(range(0, num_timesteps, i))
+            raise ValueError(
+                f"cannot create exactly {num_timesteps} steps with an integer stride"
+            )
+        section_counts = [int(x) for x in section_counts.split(",")]
+    size_per = num_timesteps // len(section_counts)
+    extra = num_timesteps % len(section_counts)
+    start_idx = 0
+    all_steps = []
+    for i, section_count in enumerate(section_counts):
+        size = size_per + (1 if i < extra else 0)
+        if size < section_count:
+            raise ValueError(
+                f"cannot divide section of {size} steps into {section_count}"
+            )
+        if section_count <= 1:
+            frac_stride = 1
+        else:
+            frac_stride = (size - 1) / (section_count - 1)
+        cur_idx = 0.0
+        taken_steps = []
+        for _ in range(section_count):
+            taken_steps.append(start_idx + round(cur_idx))
+            cur_idx += frac_stride
+        all_steps += taken_steps
+        start_idx += size
+    return set(all_steps)
+
+
+class SpacedDiffusion(GaussianDiffusion):
+    """
+    A diffusion process which can skip steps in a base diffusion process.
+
+    :param use_timesteps: a collection (sequence or set) of timesteps from the
+                          original diffusion process to retain.
+    :param kwargs: the kwargs to create the base diffusion process.
+    """
+
+    def __init__(self, use_timesteps, **kwargs):
+        self.use_timesteps = set(use_timesteps)
+        self.timestep_map = []
+        self.original_num_steps = len(kwargs["betas"])
+
+        base_diffusion = GaussianDiffusion(**kwargs)  # pylint: disable=missing-kwoa
+        last_alpha_cumprod = 1.0
+        new_betas = []
+        for i, alpha_cumprod in enumerate(base_diffusion.alphas_cumprod):
+            if i in self.use_timesteps:
+                new_betas.append(1 - alpha_cumprod / last_alpha_cumprod)
+                last_alpha_cumprod = alpha_cumprod
+                self.timestep_map.append(i)
+        kwargs["betas"] = np.array(new_betas)
+        super().__init__(**kwargs)
+
+    def p_mean_variance(
+        self, model, *args, **kwargs
+    ):  # pylint: disable=signature-differs
+        return super().p_mean_variance(self._wrap_model(model), *args, **kwargs)
+
+    def training_losses(
+        self, model, *args, **kwargs
+    ):  # pylint: disable=signature-differs
+        return super().training_losses(self._wrap_model(model), *args, **kwargs)
+
+    def condition_mean(self, cond_fn, *args, **kwargs):
+        return super().condition_mean(self._wrap_model(cond_fn), *args, **kwargs)
+
+    def condition_score(self, cond_fn, *args, **kwargs):
+        return super().condition_score(self._wrap_model(cond_fn), *args, **kwargs)
+
+    def _wrap_model(self, model):
+        if isinstance(model, _WrappedModel):
+            return model
+        return _WrappedModel(
+            model, self.timestep_map, self.rescale_timesteps, self.original_num_steps
+        )
+
+    def _scale_timesteps(self, t):
+        # Scaling is done by the wrapped model.
+        return t
+
+
+class _WrappedModel:
+    def __init__(self, model, timestep_map, rescale_timesteps, original_num_steps):
+        self.model = model
+        self.timestep_map = timestep_map
+        self.rescale_timesteps = rescale_timesteps
+        self.original_num_steps = original_num_steps
+
+    def __call__(self, x, ts, **kwargs):
+        map_tensor = th.tensor(self.timestep_map, device=ts.device, dtype=ts.dtype)
+        new_ts = map_tensor[ts]
+        if self.rescale_timesteps:
+            new_ts = new_ts.float() * (1000.0 / self.original_num_steps)
+        return self.model(x, new_ts, **kwargs)

guided_diffusion/guided_diffusion/script_util.py

@@ -0,0 +1,452 @@
+import argparse
+import inspect
+
+from . import gaussian_diffusion as gd
+from .respace import SpacedDiffusion, space_timesteps
+from .unet import SuperResModel, UNetModel, EncoderUNetModel
+
+NUM_CLASSES = 1000
+
+
+def diffusion_defaults():
+    """
+    Defaults for image and classifier training.
+    """
+    return dict(
+        learn_sigma=False,
+        diffusion_steps=1000,
+        noise_schedule="linear",
+        timestep_respacing="",
+        use_kl=False,
+        predict_xstart=False,
+        rescale_timesteps=False,
+        rescale_learned_sigmas=False,
+    )
+
+
+def classifier_defaults():
+    """
+    Defaults for classifier models.
+    """
+    return dict(
+        image_size=256,#64,
+        classifier_use_fp16=False,
+        classifier_width=128,
+        classifier_depth=2,
+        classifier_attention_resolutions="32,16,8",  # 16
+        classifier_use_scale_shift_norm=True,  # False
+        classifier_resblock_updown=True,  # False
+        classifier_pool="attention",
+    )
+
+
+def model_and_diffusion_defaults():
+    """
+    Defaults for image training.
+    """
+    res = dict(
+        image_size=64,
+        num_channels=128,
+        num_res_blocks=2,
+        num_heads=4,
+        num_heads_upsample=-1,
+        num_head_channels=-1,
+        attention_resolutions="16,8",
+        channel_mult="",
+        dropout=0.0,
+        class_cond=False,
+        use_checkpoint=False,
+        use_scale_shift_norm=True,
+        resblock_updown=False,
+        use_fp16=False,
+        use_new_attention_order=False,
+    )
+    res.update(diffusion_defaults())
+    return res
+
+
+def classifier_and_diffusion_defaults():
+    res = classifier_defaults()
+    res.update(diffusion_defaults())
+    return res
+
+
+def create_model_and_diffusion(
+    image_size,
+    class_cond,
+    learn_sigma,
+    num_channels,
+    num_res_blocks,
+    channel_mult,
+    num_heads,
+    num_head_channels,
+    num_heads_upsample,
+    attention_resolutions,
+    dropout,
+    diffusion_steps,
+    noise_schedule,
+    timestep_respacing,
+    use_kl,
+    predict_xstart,
+    rescale_timesteps,
+    rescale_learned_sigmas,
+    use_checkpoint,
+    use_scale_shift_norm,
+    resblock_updown,
+    use_fp16,
+    use_new_attention_order,
+):
+    model = create_model(
+        image_size,
+        num_channels,
+        num_res_blocks,
+        channel_mult=channel_mult,
+        learn_sigma=learn_sigma,
+        class_cond=class_cond,
+        use_checkpoint=use_checkpoint,
+        attention_resolutions=attention_resolutions,
+        num_heads=num_heads,
+        num_head_channels=num_head_channels,
+        num_heads_upsample=num_heads_upsample,
+        use_scale_shift_norm=use_scale_shift_norm,
+        dropout=dropout,
+        resblock_updown=resblock_updown,
+        use_fp16=use_fp16,
+        use_new_attention_order=use_new_attention_order,
+    )
+    diffusion = create_gaussian_diffusion(
+        steps=diffusion_steps,
+        learn_sigma=learn_sigma,
+        noise_schedule=noise_schedule,
+        use_kl=use_kl,
+        predict_xstart=predict_xstart,
+        rescale_timesteps=rescale_timesteps,
+        rescale_learned_sigmas=rescale_learned_sigmas,
+        timestep_respacing=timestep_respacing,
+    )
+    return model, diffusion
+
+
+def create_model(
+    image_size,
+    num_channels,
+    num_res_blocks,
+    channel_mult="",
+    learn_sigma=False,
+    class_cond=False,
+    use_checkpoint=False,
+    attention_resolutions="16",
+    num_heads=1,
+    num_head_channels=-1,
+    num_heads_upsample=-1,
+    use_scale_shift_norm=False,
+    dropout=0,
+    resblock_updown=False,
+    use_fp16=False,
+    use_new_attention_order=False,
+):
+    if channel_mult == "":
+        if image_size == 512:
+            channel_mult = (0.5, 1, 1, 2, 2, 4, 4)
+        elif image_size == 256:
+            channel_mult = (1, 1, 2, 2, 4, 4)
+        elif image_size == 128:
+            channel_mult = (1, 1, 2, 3, 4)
+        elif image_size == 64:
+            channel_mult = (1, 2, 3, 4)
+        else:
+            raise ValueError(f"unsupported image size: {image_size}")
+    else:
+        channel_mult = tuple(int(ch_mult) for ch_mult in channel_mult.split(","))
+
+    attention_ds = []
+    for res in attention_resolutions.split(","):
+        attention_ds.append(image_size // int(res))
+
+    return UNetModel(
+        image_size=image_size,
+        in_channels=3,
+        model_channels=num_channels,
+        out_channels=(3 if not learn_sigma else 6),
+        num_res_blocks=num_res_blocks,
+        attention_resolutions=tuple(attention_ds),
+        dropout=dropout,
+        channel_mult=channel_mult,
+        num_classes=(NUM_CLASSES if class_cond else None),
+        use_checkpoint=use_checkpoint,
+        use_fp16=use_fp16,
+        num_heads=num_heads,
+        num_head_channels=num_head_channels,
+        num_heads_upsample=num_heads_upsample,
+        use_scale_shift_norm=use_scale_shift_norm,
+        resblock_updown=resblock_updown,
+        use_new_attention_order=use_new_attention_order,
+    )
+
+
+def create_classifier_and_diffusion(
+    image_size,
+    classifier_use_fp16,
+    classifier_width,
+    classifier_depth,
+    classifier_attention_resolutions,
+    classifier_use_scale_shift_norm,
+    classifier_resblock_updown,
+    classifier_pool,
+    learn_sigma,
+    diffusion_steps,
+    noise_schedule,
+    timestep_respacing,
+    use_kl,
+    predict_xstart,
+    rescale_timesteps,
+    rescale_learned_sigmas,
+):
+    classifier = create_classifier(
+        image_size,
+        classifier_use_fp16,
+        classifier_width,
+        classifier_depth,
+        classifier_attention_resolutions,
+        classifier_use_scale_shift_norm,
+        classifier_resblock_updown,
+        classifier_pool,
+    )
+    diffusion = create_gaussian_diffusion(
+        steps=diffusion_steps,
+        learn_sigma=learn_sigma,
+        noise_schedule=noise_schedule,
+        use_kl=use_kl,
+        predict_xstart=predict_xstart,
+        rescale_timesteps=rescale_timesteps,
+        rescale_learned_sigmas=rescale_learned_sigmas,
+        timestep_respacing=timestep_respacing,
+    )
+    return classifier, diffusion
+
+
+def create_classifier(
+    image_size,
+    classifier_use_fp16,
+    classifier_width,
+    classifier_depth,
+    classifier_attention_resolutions,
+    classifier_use_scale_shift_norm,
+    classifier_resblock_updown,
+    classifier_pool,
+):
+    if image_size == 512:
+        channel_mult = (0.5, 1, 1, 2, 2, 4, 4)
+    elif image_size == 256:
+        channel_mult = (1, 1, 2, 2, 4, 4)
+    elif image_size == 128:
+        channel_mult = (1, 1, 2, 3, 4)
+    elif image_size == 64:
+        channel_mult = (1, 2, 3, 4)
+    else:
+        raise ValueError(f"unsupported image size: {image_size}")
+
+    attention_ds = []
+    for res in classifier_attention_resolutions.split(","):
+        attention_ds.append(image_size // int(res))
+
+    return EncoderUNetModel(
+        image_size=image_size,
+        in_channels=3,
+        model_channels=classifier_width,
+        out_channels=1000,
+        num_res_blocks=classifier_depth,
+        attention_resolutions=tuple(attention_ds),
+        channel_mult=channel_mult,
+        use_fp16=classifier_use_fp16,
+        num_head_channels=64,
+        use_scale_shift_norm=classifier_use_scale_shift_norm,
+        resblock_updown=classifier_resblock_updown,
+        pool=classifier_pool,
+    )
+
+
+def sr_model_and_diffusion_defaults():
+    res = model_and_diffusion_defaults()
+    res["large_size"] = 256
+    res["small_size"] = 64
+    arg_names = inspect.getfullargspec(sr_create_model_and_diffusion)[0]
+    for k in res.copy().keys():
+        if k not in arg_names:
+            del res[k]
+    return res
+
+
+def sr_create_model_and_diffusion(
+    large_size,
+    small_size,
+    class_cond,
+    learn_sigma,
+    num_channels,
+    num_res_blocks,
+    num_heads,
+    num_head_channels,
+    num_heads_upsample,
+    attention_resolutions,
+    dropout,
+    diffusion_steps,
+    noise_schedule,
+    timestep_respacing,
+    use_kl,
+    predict_xstart,
+    rescale_timesteps,
+    rescale_learned_sigmas,
+    use_checkpoint,
+    use_scale_shift_norm,
+    resblock_updown,
+    use_fp16,
+):
+    model = sr_create_model(
+        large_size,
+        small_size,
+        num_channels,
+        num_res_blocks,
+        learn_sigma=learn_sigma,
+        class_cond=class_cond,
+        use_checkpoint=use_checkpoint,
+        attention_resolutions=attention_resolutions,
+        num_heads=num_heads,
+        num_head_channels=num_head_channels,
+        num_heads_upsample=num_heads_upsample,
+        use_scale_shift_norm=use_scale_shift_norm,
+        dropout=dropout,
+        resblock_updown=resblock_updown,
+        use_fp16=use_fp16,
+    )
+    diffusion = create_gaussian_diffusion(
+        steps=diffusion_steps,
+        learn_sigma=learn_sigma,
+        noise_schedule=noise_schedule,
+        use_kl=use_kl,
+        predict_xstart=predict_xstart,
+        rescale_timesteps=rescale_timesteps,
+        rescale_learned_sigmas=rescale_learned_sigmas,
+        timestep_respacing=timestep_respacing,
+    )
+    return model, diffusion
+
+
+def sr_create_model(
+    large_size,
+    small_size,
+    num_channels,
+    num_res_blocks,
+    learn_sigma,
+    class_cond,
+    use_checkpoint,
+    attention_resolutions,
+    num_heads,
+    num_head_channels,
+    num_heads_upsample,
+    use_scale_shift_norm,
+    dropout,
+    resblock_updown,
+    use_fp16,
+):
+    _ = small_size  # hack to prevent unused variable
+
+    if large_size == 512:
+        channel_mult = (1, 1, 2, 2, 4, 4)
+    elif large_size == 256:
+        channel_mult = (1, 1, 2, 2, 4, 4)
+    elif large_size == 64:
+        channel_mult = (1, 2, 3, 4)
+    else:
+        raise ValueError(f"unsupported large size: {large_size}")
+
+    attention_ds = []
+    for res in attention_resolutions.split(","):
+        attention_ds.append(large_size // int(res))
+
+    return SuperResModel(
+        image_size=large_size,
+        in_channels=3,
+        model_channels=num_channels,
+        out_channels=(3 if not learn_sigma else 6),
+        num_res_blocks=num_res_blocks,
+        attention_resolutions=tuple(attention_ds),
+        dropout=dropout,
+        channel_mult=channel_mult,
+        num_classes=(NUM_CLASSES if class_cond else None),
+        use_checkpoint=use_checkpoint,
+        num_heads=num_heads,
+        num_head_channels=num_head_channels,
+        num_heads_upsample=num_heads_upsample,
+        use_scale_shift_norm=use_scale_shift_norm,
+        resblock_updown=resblock_updown,
+        use_fp16=use_fp16,
+    )
+
+
+def create_gaussian_diffusion(
+    *,
+    steps=1000,
+    learn_sigma=False,
+    sigma_small=False,
+    noise_schedule="linear",
+    use_kl=False,
+    predict_xstart=False,
+    rescale_timesteps=False,
+    rescale_learned_sigmas=False,
+    timestep_respacing="",
+):
+    betas = gd.get_named_beta_schedule(noise_schedule, steps)
+    if use_kl:
+        loss_type = gd.LossType.RESCALED_KL
+    elif rescale_learned_sigmas:
+        loss_type = gd.LossType.RESCALED_MSE
+    else:
+        loss_type = gd.LossType.MSE
+    if not timestep_respacing:
+        timestep_respacing = [steps]
+    return SpacedDiffusion(
+        use_timesteps=space_timesteps(steps, timestep_respacing),
+        betas=betas,
+        model_mean_type=(
+            gd.ModelMeanType.EPSILON if not predict_xstart else gd.ModelMeanType.START_X
+        ),
+        model_var_type=(
+            (
+                gd.ModelVarType.FIXED_LARGE
+                if not sigma_small
+                else gd.ModelVarType.FIXED_SMALL
+            )
+            if not learn_sigma
+            else gd.ModelVarType.LEARNED_RANGE
+        ),
+        loss_type=loss_type,
+        rescale_timesteps=rescale_timesteps,
+    )
+
+
+def add_dict_to_argparser(parser, default_dict):
+    for k, v in default_dict.items():
+        v_type = type(v)
+        if v is None:
+            v_type = str
+        elif isinstance(v, bool):
+            v_type = str2bool
+        parser.add_argument(f"--{k}", default=v, type=v_type)
+
+
+def args_to_dict(args, keys):
+    return {k: getattr(args, k) for k in keys}
+
+
+def str2bool(v):
+    """
+    https://stackoverflow.com/questions/15008758/parsing-boolean-values-with-argparse
+    """
+    if isinstance(v, bool):
+        return v
+    if v.lower() in ("yes", "true", "t", "y", "1"):
+        return True
+    elif v.lower() in ("no", "false", "f", "n", "0"):
+        return False
+    else:
+        raise argparse.ArgumentTypeError("boolean value expected")

guided_diffusion/guided_diffusion/train_util.py

@@ -0,0 +1,301 @@
+import copy
+import functools
+import os
+
+import blobfile as bf
+import torch as th
+import torch.distributed as dist
+from torch.nn.parallel.distributed import DistributedDataParallel as DDP
+from torch.optim import AdamW
+
+from . import dist_util, logger
+from .fp16_util import MixedPrecisionTrainer
+from .nn import update_ema
+from .resample import LossAwareSampler, UniformSampler
+
+# For ImageNet experiments, this was a good default value.
+# We found that the lg_loss_scale quickly climbed to
+# 20-21 within the first ~1K steps of training.
+INITIAL_LOG_LOSS_SCALE = 20.0
+
+
+class TrainLoop:
+    def __init__(
+        self,
+        *,
+        model,
+        diffusion,
+        data,
+        batch_size,
+        microbatch,
+        lr,
+        ema_rate,
+        log_interval,
+        save_interval,
+        resume_checkpoint,
+        use_fp16=False,
+        fp16_scale_growth=1e-3,
+        schedule_sampler=None,
+        weight_decay=0.0,
+        lr_anneal_steps=0,
+    ):
+        self.model = model
+        self.diffusion = diffusion
+        self.data = data
+        self.batch_size = batch_size
+        self.microbatch = microbatch if microbatch > 0 else batch_size
+        self.lr = lr
+        self.ema_rate = (
+            [ema_rate]
+            if isinstance(ema_rate, float)
+            else [float(x) for x in ema_rate.split(",")]
+        )
+        self.log_interval = log_interval
+        self.save_interval = save_interval
+        self.resume_checkpoint = resume_checkpoint
+        self.use_fp16 = use_fp16
+        self.fp16_scale_growth = fp16_scale_growth
+        self.schedule_sampler = schedule_sampler or UniformSampler(diffusion)
+        self.weight_decay = weight_decay
+        self.lr_anneal_steps = lr_anneal_steps
+
+        self.step = 0
+        self.resume_step = 0
+        self.global_batch = self.batch_size * dist.get_world_size()
+
+        self.sync_cuda = th.cuda.is_available()
+
+        self._load_and_sync_parameters()
+        self.mp_trainer = MixedPrecisionTrainer(
+            model=self.model,
+            use_fp16=self.use_fp16,
+            fp16_scale_growth=fp16_scale_growth,
+        )
+
+        self.opt = AdamW(
+            self.mp_trainer.master_params, lr=self.lr, weight_decay=self.weight_decay
+        )
+        if self.resume_step:
+            self._load_optimizer_state()
+            # Model was resumed, either due to a restart or a checkpoint
+            # being specified at the command line.
+            self.ema_params = [
+                self._load_ema_parameters(rate) for rate in self.ema_rate
+            ]
+        else:
+            self.ema_params = [
+                copy.deepcopy(self.mp_trainer.master_params)
+                for _ in range(len(self.ema_rate))
+            ]
+
+        if th.cuda.is_available():
+            self.use_ddp = True
+            self.ddp_model = DDP(
+                self.model,
+                device_ids=[dist_util.dev()],
+                output_device=dist_util.dev(),
+                broadcast_buffers=False,
+                bucket_cap_mb=128,
+                find_unused_parameters=False,
+            )
+        else:
+            if dist.get_world_size() > 1:
+                logger.warn(
+                    "Distributed training requires CUDA. "
+                    "Gradients will not be synchronized properly!"
+                )
+            self.use_ddp = False
+            self.ddp_model = self.model
+
+    def _load_and_sync_parameters(self):
+        resume_checkpoint = find_resume_checkpoint() or self.resume_checkpoint
+
+        if resume_checkpoint:
+            self.resume_step = parse_resume_step_from_filename(resume_checkpoint)
+            if dist.get_rank() == 0:
+                logger.log(f"loading model from checkpoint: {resume_checkpoint}...")
+                self.model.load_state_dict(
+                    dist_util.load_state_dict(
+                        resume_checkpoint, map_location=dist_util.dev()
+                    )
+                )
+
+        dist_util.sync_params(self.model.parameters())
+
+    def _load_ema_parameters(self, rate):
+        ema_params = copy.deepcopy(self.mp_trainer.master_params)
+
+        main_checkpoint = find_resume_checkpoint() or self.resume_checkpoint
+        ema_checkpoint = find_ema_checkpoint(main_checkpoint, self.resume_step, rate)
+        if ema_checkpoint:
+            if dist.get_rank() == 0:
+                logger.log(f"loading EMA from checkpoint: {ema_checkpoint}...")
+                state_dict = dist_util.load_state_dict(
+                    ema_checkpoint, map_location=dist_util.dev()
+                )
+                ema_params = self.mp_trainer.state_dict_to_master_params(state_dict)
+
+        dist_util.sync_params(ema_params)
+        return ema_params
+
+    def _load_optimizer_state(self):
+        main_checkpoint = find_resume_checkpoint() or self.resume_checkpoint
+        opt_checkpoint = bf.join(
+            bf.dirname(main_checkpoint), f"opt{self.resume_step:06}.pt"
+        )
+        if bf.exists(opt_checkpoint):
+            logger.log(f"loading optimizer state from checkpoint: {opt_checkpoint}")
+            state_dict = dist_util.load_state_dict(
+                opt_checkpoint, map_location=dist_util.dev()
+            )
+            self.opt.load_state_dict(state_dict)
+
+    def run_loop(self):
+        while (
+            not self.lr_anneal_steps
+            or self.step + self.resume_step < self.lr_anneal_steps
+        ):
+            batch, cond = next(self.data)
+            self.run_step(batch, cond)
+            if self.step % self.log_interval == 0:
+                logger.dumpkvs()
+            if self.step % self.save_interval == 0:
+                self.save()
+                # Run for a finite amount of time in integration tests.
+                if os.environ.get("DIFFUSION_TRAINING_TEST", "") and self.step > 0:
+                    return
+            self.step += 1
+        # Save the last checkpoint if it wasn't already saved.
+        if (self.step - 1) % self.save_interval != 0:
+            self.save()
+
+    def run_step(self, batch, cond):
+        self.forward_backward(batch, cond)
+        took_step = self.mp_trainer.optimize(self.opt)
+        if took_step:
+            self._update_ema()
+        self._anneal_lr()
+        self.log_step()
+
+    def forward_backward(self, batch, cond):
+        self.mp_trainer.zero_grad()
+        for i in range(0, batch.shape[0], self.microbatch):
+            micro = batch[i : i + self.microbatch].to(dist_util.dev())
+            micro_cond = {
+                k: v[i : i + self.microbatch].to(dist_util.dev())
+                for k, v in cond.items()
+            }
+            last_batch = (i + self.microbatch) >= batch.shape[0]
+            t, weights = self.schedule_sampler.sample(micro.shape[0], dist_util.dev())
+
+            compute_losses = functools.partial(
+                self.diffusion.training_losses,
+                self.ddp_model,
+                micro,
+                t,
+                model_kwargs=micro_cond,
+            )
+
+            if last_batch or not self.use_ddp:
+                losses = compute_losses()
+            else:
+                with self.ddp_model.no_sync():
+                    losses = compute_losses()
+
+            if isinstance(self.schedule_sampler, LossAwareSampler):
+                self.schedule_sampler.update_with_local_losses(
+                    t, losses["loss"].detach()
+                )
+
+            loss = (losses["loss"] * weights).mean()
+            log_loss_dict(
+                self.diffusion, t, {k: v * weights for k, v in losses.items()}
+            )
+            self.mp_trainer.backward(loss)
+
+    def _update_ema(self):
+        for rate, params in zip(self.ema_rate, self.ema_params):
+            update_ema(params, self.mp_trainer.master_params, rate=rate)
+
+    def _anneal_lr(self):
+        if not self.lr_anneal_steps:
+            return
+        frac_done = (self.step + self.resume_step) / self.lr_anneal_steps
+        lr = self.lr * (1 - frac_done)
+        for param_group in self.opt.param_groups:
+            param_group["lr"] = lr
+
+    def log_step(self):
+        logger.logkv("step", self.step + self.resume_step)
+        logger.logkv("samples", (self.step + self.resume_step + 1) * self.global_batch)
+
+    def save(self):
+        def save_checkpoint(rate, params):
+            state_dict = self.mp_trainer.master_params_to_state_dict(params)
+            if dist.get_rank() == 0:
+                logger.log(f"saving model {rate}...")
+                if not rate:
+                    filename = f"model{(self.step+self.resume_step):06d}.pt"
+                else:
+                    filename = f"ema_{rate}_{(self.step+self.resume_step):06d}.pt"
+                with bf.BlobFile(bf.join(get_blob_logdir(), filename), "wb") as f:
+                    th.save(state_dict, f)
+
+        save_checkpoint(0, self.mp_trainer.master_params)
+        for rate, params in zip(self.ema_rate, self.ema_params):
+            save_checkpoint(rate, params)
+
+        if dist.get_rank() == 0:
+            with bf.BlobFile(
+                bf.join(get_blob_logdir(), f"opt{(self.step+self.resume_step):06d}.pt"),
+                "wb",
+            ) as f:
+                th.save(self.opt.state_dict(), f)
+
+        dist.barrier()
+
+
+def parse_resume_step_from_filename(filename):
+    """
+    Parse filenames of the form path/to/modelNNNNNN.pt, where NNNNNN is the
+    checkpoint's number of steps.
+    """
+    split = filename.split("model")
+    if len(split) < 2:
+        return 0
+    split1 = split[-1].split(".")[0]
+    try:
+        return int(split1)
+    except ValueError:
+        return 0
+
+
+def get_blob_logdir():
+    # You can change this to be a separate path to save checkpoints to
+    # a blobstore or some external drive.
+    return logger.get_dir()
+
+
+def find_resume_checkpoint():
+    # On your infrastructure, you may want to override this to automatically
+    # discover the latest checkpoint on your blob storage, etc.
+    return None
+
+
+def find_ema_checkpoint(main_checkpoint, step, rate):
+    if main_checkpoint is None:
+        return None
+    filename = f"ema_{rate}_{(step):06d}.pt"
+    path = bf.join(bf.dirname(main_checkpoint), filename)
+    if bf.exists(path):
+        return path
+    return None
+
+
+def log_loss_dict(diffusion, ts, losses):
+    for key, values in losses.items():
+        logger.logkv_mean(key, values.mean().item())
+        # Log the quantiles (four quartiles, in particular).
+        for sub_t, sub_loss in zip(ts.cpu().numpy(), values.detach().cpu().numpy()):
+            quartile = int(4 * sub_t / diffusion.num_timesteps)
+            logger.logkv_mean(f"{key}_q{quartile}", sub_loss)

guided_diffusion/guided_diffusion/unet.py

@@ -0,0 +1,908 @@
+from abc import abstractmethod
+
+import math
+
+import numpy as np
+import torch as th
+import torch.nn as nn
+import torch.nn.functional as F
+
+from .fp16_util import convert_module_to_f16, convert_module_to_f32
+from .nn import (
+    checkpoint,
+    conv_nd,
+    linear,
+    avg_pool_nd,
+    zero_module,
+    normalization,
+    timestep_embedding,
+)
+
+
+class AttentionPool2d(nn.Module):
+    """
+    Adapted from CLIP: https://github.com/openai/CLIP/blob/main/clip/model.py
+    """
+
+    def __init__(
+        self,
+        spacial_dim: int,
+        embed_dim: int,
+        num_heads_channels: int,
+        output_dim: int = None,
+    ):
+        super().__init__()
+        self.positional_embedding = nn.Parameter(
+            th.randn(embed_dim, spacial_dim ** 2 + 1) / embed_dim ** 0.5
+        )
+        self.qkv_proj = conv_nd(1, embed_dim, 3 * embed_dim, 1)
+        self.c_proj = conv_nd(1, embed_dim, output_dim or embed_dim, 1)
+        self.num_heads = embed_dim // num_heads_channels
+        self.attention = QKVAttention(self.num_heads)
+
+    def forward(self, x):
+        b, c, *_spatial = x.shape
+        x = x.reshape(b, c, -1)  # NC(HW)
+        x = th.cat([x.mean(dim=-1, keepdim=True), x], dim=-1)  # NC(HW+1)
+        x = x + self.positional_embedding[None, :, :].to(x.dtype)  # NC(HW+1)
+        x = self.qkv_proj(x)
+        x = self.attention(x)
+        x = self.c_proj(x)
+        return x[:, :, 0]
+
+
+class TimestepBlock(nn.Module):
+    """
+    Any module where forward() takes timestep embeddings as a second argument.
+    """
+
+    @abstractmethod
+    def forward(self, x, emb):
+        """
+        Apply the module to `x` given `emb` timestep embeddings.
+        """
+
+
+class TimestepEmbedSequential(nn.Sequential, TimestepBlock):
+    """
+    A sequential module that passes timestep embeddings to the children that
+    support it as an extra input.
+    """
+
+    def forward(self, x, emb):
+        for layer in self:
+            if isinstance(layer, TimestepBlock):
+                x = layer(x, emb)
+            else:
+                x = layer(x)
+        return x
+
+
+class Upsample(nn.Module):
+    """
+    An upsampling layer with an optional convolution.
+
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 upsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        if use_conv:
+            self.conv = conv_nd(dims, self.channels, self.out_channels, 3, padding=1)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        if self.dims == 3:
+            x = F.interpolate(
+                x, (x.shape[2], x.shape[3] * 2, x.shape[4] * 2), mode="nearest"
+            )
+        else:
+            x = F.interpolate(x, scale_factor=2, mode="nearest")
+        if self.use_conv:
+            x = self.conv(x)
+        return x
+
+
+class Downsample(nn.Module):
+    """
+    A downsampling layer with an optional convolution.
+
+    :param channels: channels in the inputs and outputs.
+    :param use_conv: a bool determining if a convolution is applied.
+    :param dims: determines if the signal is 1D, 2D, or 3D. If 3D, then
+                 downsampling occurs in the inner-two dimensions.
+    """
+
+    def __init__(self, channels, use_conv, dims=2, out_channels=None):
+        super().__init__()
+        self.channels = channels
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.dims = dims
+        stride = 2 if dims != 3 else (1, 2, 2)
+        if use_conv:
+            self.op = conv_nd(
+                dims, self.channels, self.out_channels, 3, stride=stride, padding=1
+            )
+        else:
+            assert self.channels == self.out_channels
+            self.op = avg_pool_nd(dims, kernel_size=stride, stride=stride)
+
+    def forward(self, x):
+        assert x.shape[1] == self.channels
+        return self.op(x)
+
+
+class ResBlock(TimestepBlock):
+    """
+    A residual block that can optionally change the number of channels.
+
+    :param channels: the number of input channels.
+    :param emb_channels: the number of timestep embedding channels.
+    :param dropout: the rate of dropout.
+    :param out_channels: if specified, the number of out channels.
+    :param use_conv: if True and out_channels is specified, use a spatial
+        convolution instead of a smaller 1x1 convolution to change the
+        channels in the skip connection.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param use_checkpoint: if True, use gradient checkpointing on this module.
+    :param up: if True, use this block for upsampling.
+    :param down: if True, use this block for downsampling.
+    """
+
+    def __init__(
+        self,
+        channels,
+        emb_channels,
+        dropout,
+        out_channels=None,
+        use_conv=False,
+        use_scale_shift_norm=False,
+        dims=2,
+        use_checkpoint=False,
+        up=False,
+        down=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.emb_channels = emb_channels
+        self.dropout = dropout
+        self.out_channels = out_channels or channels
+        self.use_conv = use_conv
+        self.use_checkpoint = use_checkpoint
+        self.use_scale_shift_norm = use_scale_shift_norm
+
+        self.in_layers = nn.Sequential(
+            normalization(channels),
+            nn.SiLU(),
+            conv_nd(dims, channels, self.out_channels, 3, padding=1),
+        )
+
+        self.updown = up or down
+
+        if up:
+            self.h_upd = Upsample(channels, False, dims)
+            self.x_upd = Upsample(channels, False, dims)
+        elif down:
+            self.h_upd = Downsample(channels, False, dims)
+            self.x_upd = Downsample(channels, False, dims)
+        else:
+            self.h_upd = self.x_upd = nn.Identity()
+
+        self.emb_layers = nn.Sequential(
+            nn.SiLU(),
+            linear(
+                emb_channels,
+                2 * self.out_channels if use_scale_shift_norm else self.out_channels,
+            ),
+        )
+        self.out_layers = nn.Sequential(
+            normalization(self.out_channels),
+            nn.SiLU(),
+            nn.Dropout(p=dropout),
+            zero_module(
+                conv_nd(dims, self.out_channels, self.out_channels, 3, padding=1)
+            ),
+        )
+
+        if self.out_channels == channels:
+            self.skip_connection = nn.Identity()
+        elif use_conv:
+            self.skip_connection = conv_nd(
+                dims, channels, self.out_channels, 3, padding=1
+            )
+        else:
+            self.skip_connection = conv_nd(dims, channels, self.out_channels, 1)
+
+    def forward(self, x, emb):
+        """
+        Apply the block to a Tensor, conditioned on a timestep embedding.
+
+        :param x: an [N x C x ...] Tensor of features.
+        :param emb: an [N x emb_channels] Tensor of timestep embeddings.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        return checkpoint(
+            self._forward, (x, emb), self.parameters(), self.use_checkpoint
+        )
+
+    def _forward(self, x, emb):
+        if self.updown:
+            in_rest, in_conv = self.in_layers[:-1], self.in_layers[-1]
+            h = in_rest(x)
+            h = self.h_upd(h)
+            x = self.x_upd(x)
+            h = in_conv(h)
+        else:
+            h = self.in_layers(x)
+        emb_out = self.emb_layers(emb).type(h.dtype)
+        while len(emb_out.shape) < len(h.shape):
+            emb_out = emb_out[..., None]
+        if self.use_scale_shift_norm:
+            out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
+            scale, shift = th.chunk(emb_out, 2, dim=1)
+            h = out_norm(h) * (1 + scale) + shift
+            h = out_rest(h)
+        else:
+            h = h + emb_out
+            h = self.out_layers(h)
+        return self.skip_connection(x) + h
+
+
+class AttentionBlock(nn.Module):
+    """
+    An attention block that allows spatial positions to attend to each other.
+
+    Originally ported from here, but adapted to the N-d case.
+    https://github.com/hojonathanho/diffusion/blob/1e0dceb3b3495bbe19116a5e1b3596cd0706c543/diffusion_tf/models/unet.py#L66.
+    """
+
+    def __init__(
+        self,
+        channels,
+        num_heads=1,
+        num_head_channels=-1,
+        use_checkpoint=False,
+        use_new_attention_order=False,
+    ):
+        super().__init__()
+        self.channels = channels
+        if num_head_channels == -1:
+            self.num_heads = num_heads
+        else:
+            assert (
+                channels % num_head_channels == 0
+            ), f"q,k,v channels {channels} is not divisible by num_head_channels {num_head_channels}"
+            self.num_heads = channels // num_head_channels
+        self.use_checkpoint = use_checkpoint
+        self.norm = normalization(channels)
+        self.qkv = conv_nd(1, channels, channels * 3, 1)
+        if use_new_attention_order:
+            # split qkv before split heads
+            self.attention = QKVAttention(self.num_heads)
+        else:
+            # split heads before split qkv
+            self.attention = QKVAttentionLegacy(self.num_heads)
+
+        self.proj_out = zero_module(conv_nd(1, channels, channels, 1))
+
+    def forward(self, x):
+        return checkpoint(self._forward, (x,), self.parameters(), True)
+
+    def _forward(self, x):
+        b, c, *spatial = x.shape
+        x = x.reshape(b, c, -1)
+        qkv = self.qkv(self.norm(x))
+        h = self.attention(qkv)
+        h = self.proj_out(h)
+        return (x + h).reshape(b, c, *spatial)
+
+
+def count_flops_attn(model, _x, y):
+    """
+    A counter for the `thop` package to count the operations in an
+    attention operation.
+    Meant to be used like:
+        macs, params = thop.profile(
+            model,
+            inputs=(inputs, timestamps),
+            custom_ops={QKVAttention: QKVAttention.count_flops},
+        )
+    """
+    b, c, *spatial = y[0].shape
+    num_spatial = int(np.prod(spatial))
+    # We perform two matmuls with the same number of ops.
+    # The first computes the weight matrix, the second computes
+    # the combination of the value vectors.
+    matmul_ops = 2 * b * (num_spatial ** 2) * c
+    model.total_ops += th.DoubleTensor([matmul_ops])
+
+
+class QKVAttentionLegacy(nn.Module):
+    """
+    A module which performs QKV attention. Matches legacy QKVAttention + input/ouput heads shaping
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+
+        :param qkv: an [N x (H * 3 * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.reshape(bs * self.n_heads, ch * 3, length).split(ch, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts", q * scale, k * scale
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v)
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class QKVAttention(nn.Module):
+    """
+    A module which performs QKV attention and splits in a different order.
+    """
+
+    def __init__(self, n_heads):
+        super().__init__()
+        self.n_heads = n_heads
+
+    def forward(self, qkv):
+        """
+        Apply QKV attention.
+
+        :param qkv: an [N x (3 * H * C) x T] tensor of Qs, Ks, and Vs.
+        :return: an [N x (H * C) x T] tensor after attention.
+        """
+        bs, width, length = qkv.shape
+        assert width % (3 * self.n_heads) == 0
+        ch = width // (3 * self.n_heads)
+        q, k, v = qkv.chunk(3, dim=1)
+        scale = 1 / math.sqrt(math.sqrt(ch))
+        weight = th.einsum(
+            "bct,bcs->bts",
+            (q * scale).view(bs * self.n_heads, ch, length),
+            (k * scale).view(bs * self.n_heads, ch, length),
+        )  # More stable with f16 than dividing afterwards
+        weight = th.softmax(weight.float(), dim=-1).type(weight.dtype)
+        a = th.einsum("bts,bcs->bct", weight, v.reshape(bs * self.n_heads, ch, length))
+        return a.reshape(bs, -1, length)
+
+    @staticmethod
+    def count_flops(model, _x, y):
+        return count_flops_attn(model, _x, y)
+
+
+class UNetModel(nn.Module):
+    """
+    The full UNet model with attention and timestep embedding.
+
+    :param in_channels: channels in the input Tensor.
+    :param model_channels: base channel count for the model.
+    :param out_channels: channels in the output Tensor.
+    :param num_res_blocks: number of residual blocks per downsample.
+    :param attention_resolutions: a collection of downsample rates at which
+        attention will take place. May be a set, list, or tuple.
+        For example, if this contains 4, then at 4x downsampling, attention
+        will be used.
+    :param dropout: the dropout probability.
+    :param channel_mult: channel multiplier for each level of the UNet.
+    :param conv_resample: if True, use learned convolutions for upsampling and
+        downsampling.
+    :param dims: determines if the signal is 1D, 2D, or 3D.
+    :param num_classes: if specified (as an int), then this model will be
+        class-conditional with `num_classes` classes.
+    :param use_checkpoint: use gradient checkpointing to reduce memory usage.
+    :param num_heads: the number of attention heads in each attention layer.
+    :param num_heads_channels: if specified, ignore num_heads and instead use
+                               a fixed channel width per attention head.
+    :param num_heads_upsample: works with num_heads to set a different number
+                               of heads for upsampling. Deprecated.
+    :param use_scale_shift_norm: use a FiLM-like conditioning mechanism.
+    :param resblock_updown: use residual blocks for up/downsampling.
+    :param use_new_attention_order: use a different attention pattern for potentially
+                                    increased efficiency.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        attention_resolutions,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        num_classes=None,
+        use_checkpoint=False,
+        use_fp16=False,
+        num_heads=1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+    ):
+        super().__init__()
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        self.image_size = image_size
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+        self.num_res_blocks = num_res_blocks
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.num_classes = num_classes
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        if self.num_classes is not None:
+            self.label_emb = nn.Embedding(num_classes, time_embed_dim)
+
+        ch = input_ch = int(channel_mult[0] * model_channels)
+        self.input_blocks = nn.ModuleList(
+            [TimestepEmbedSequential(conv_nd(dims, in_channels, ch, 3, padding=1))]
+        )
+        self._feature_size = ch
+        input_block_chans = [ch]
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for _ in range(num_res_blocks):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=int(mult * model_channels),
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = int(mult * model_channels)
+                if ds in attention_resolutions:
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads,
+                            num_head_channels=num_head_channels,
+                            use_new_attention_order=use_new_attention_order,
+                        )
+                    )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=num_head_channels,
+                use_new_attention_order=use_new_attention_order,
+            ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self._feature_size += ch
+
+        self.output_blocks = nn.ModuleList([])
+        for level, mult in list(enumerate(channel_mult))[::-1]:
+            for i in range(num_res_blocks + 1):
+                ich = input_block_chans.pop()
+                layers = [
+                    ResBlock(
+                        ch + ich,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=int(model_channels * mult),
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = int(model_channels * mult)
+                if ds in attention_resolutions:
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads_upsample,
+                            num_head_channels=num_head_channels,
+                            use_new_attention_order=use_new_attention_order,
+                        )
+                    )
+                if level and i == num_res_blocks:
+                    out_ch = ch
+                    layers.append(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            up=True,
+                        )
+                        if resblock_updown
+                        else Upsample(ch, conv_resample, dims=dims, out_channels=out_ch)
+                    )
+                    ds //= 2
+                self.output_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+
+        self.out = nn.Sequential(
+            normalization(ch),
+            nn.SiLU(),
+            zero_module(conv_nd(dims, input_ch, out_channels, 3, padding=1)),
+        )
+
+    def convert_to_fp16(self):
+        """
+        Convert the torso of the model to float16.
+        """
+        self.input_blocks.apply(convert_module_to_f16)
+        self.middle_block.apply(convert_module_to_f16)
+        self.output_blocks.apply(convert_module_to_f16)
+
+    def convert_to_fp32(self):
+        """
+        Convert the torso of the model to float32.
+        """
+        self.input_blocks.apply(convert_module_to_f32)
+        self.middle_block.apply(convert_module_to_f32)
+        self.output_blocks.apply(convert_module_to_f32)
+
+    def forward(self, x, timesteps, middle_out=False, y=None):
+        """
+        Apply the model to an input batch.
+
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :param y: an [N] Tensor of labels, if class-conditional.
+        :return: an [N x C x ...] Tensor of outputs.
+        """
+        #assert (y is not None) == (
+        #    self.num_classes is not None
+        #), "must specify y if and only if the model is class-conditional"
+
+        hs = []
+        emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
+
+        if self.num_classes is not None:
+            try:
+                assert y.shape == (x.shape[0],)
+                emb = emb + self.label_emb(y)
+            except Exception as err:
+                print(str(err))
+
+        h = x.type(self.dtype)
+        for module in self.input_blocks:
+            h = module(h, emb)
+            hs.append(h)
+        h = self.middle_block(h, emb)
+
+        if middle_out:
+            middle_out_emb = h.clone()
+        else:
+            middle_out_emb = None
+
+        for module in self.output_blocks:
+            h = th.cat([h, hs.pop()], dim=1)
+            h = module(h, emb)
+        h = h.type(x.dtype)
+
+        if middle_out:
+            return self.out(h), middle_out_emb
+        else:
+            return self.out(h)
+
+
+
+class SuperResModel(UNetModel):
+    """
+    A UNetModel that performs super-resolution.
+
+    Expects an extra kwarg `low_res` to condition on a low-resolution image.
+    """
+
+    def __init__(self, image_size, in_channels, *args, **kwargs):
+        super().__init__(image_size, in_channels * 2, *args, **kwargs)
+
+    def forward(self, x, timesteps, low_res=None, **kwargs):
+        _, _, new_height, new_width = x.shape
+        upsampled = F.interpolate(low_res, (new_height, new_width), mode="bilinear")
+        x = th.cat([x, upsampled], dim=1)
+        return super().forward(x, timesteps, **kwargs)
+
+
+class EncoderUNetModel(nn.Module):
+    """
+    The half UNet model with attention and timestep embedding.
+
+    For usage, see UNet.
+    """
+
+    def __init__(
+        self,
+        image_size,
+        in_channels,
+        model_channels,
+        out_channels,
+        num_res_blocks,
+        attention_resolutions,
+        dropout=0,
+        channel_mult=(1, 2, 4, 8),
+        conv_resample=True,
+        dims=2,
+        use_checkpoint=False,
+        use_fp16=False,
+        num_heads=1,
+        num_head_channels=-1,
+        num_heads_upsample=-1,
+        use_scale_shift_norm=False,
+        resblock_updown=False,
+        use_new_attention_order=False,
+        pool="adaptive",
+    ):
+        super().__init__()
+
+        if num_heads_upsample == -1:
+            num_heads_upsample = num_heads
+
+        self.in_channels = in_channels
+        self.model_channels = model_channels
+        self.out_channels = out_channels
+        self.num_res_blocks = num_res_blocks
+        self.attention_resolutions = attention_resolutions
+        self.dropout = dropout
+        self.channel_mult = channel_mult
+        self.conv_resample = conv_resample
+        self.use_checkpoint = use_checkpoint
+        self.dtype = th.float16 if use_fp16 else th.float32
+        self.num_heads = num_heads
+        self.num_head_channels = num_head_channels
+        self.num_heads_upsample = num_heads_upsample
+
+        time_embed_dim = model_channels * 4
+        self.time_embed = nn.Sequential(
+            linear(model_channels, time_embed_dim),
+            nn.SiLU(),
+            linear(time_embed_dim, time_embed_dim),
+        )
+
+        ch = int(channel_mult[0] * model_channels)
+        self.input_blocks = nn.ModuleList(
+            [TimestepEmbedSequential(conv_nd(dims, in_channels, ch, 3, padding=1))]
+        )
+        self._feature_size = ch
+        input_block_chans = [ch]
+        ds = 1
+        for level, mult in enumerate(channel_mult):
+            for _ in range(num_res_blocks):
+                layers = [
+                    ResBlock(
+                        ch,
+                        time_embed_dim,
+                        dropout,
+                        out_channels=int(mult * model_channels),
+                        dims=dims,
+                        use_checkpoint=use_checkpoint,
+                        use_scale_shift_norm=use_scale_shift_norm,
+                    )
+                ]
+                ch = int(mult * model_channels)
+                if ds in attention_resolutions:
+                    layers.append(
+                        AttentionBlock(
+                            ch,
+                            use_checkpoint=use_checkpoint,
+                            num_heads=num_heads,
+                            num_head_channels=num_head_channels,
+                            use_new_attention_order=use_new_attention_order,
+                        )
+                    )
+                self.input_blocks.append(TimestepEmbedSequential(*layers))
+                self._feature_size += ch
+                input_block_chans.append(ch)
+            if level != len(channel_mult) - 1:
+                out_ch = ch
+                self.input_blocks.append(
+                    TimestepEmbedSequential(
+                        ResBlock(
+                            ch,
+                            time_embed_dim,
+                            dropout,
+                            out_channels=out_ch,
+                            dims=dims,
+                            use_checkpoint=use_checkpoint,
+                            use_scale_shift_norm=use_scale_shift_norm,
+                            down=True,
+                        )
+                        if resblock_updown
+                        else Downsample(
+                            ch, conv_resample, dims=dims, out_channels=out_ch
+                        )
+                    )
+                )
+                ch = out_ch
+                input_block_chans.append(ch)
+                ds *= 2
+                self._feature_size += ch
+
+        self.middle_block = TimestepEmbedSequential(
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+            AttentionBlock(
+                ch,
+                use_checkpoint=use_checkpoint,
+                num_heads=num_heads,
+                num_head_channels=num_head_channels,
+                use_new_attention_order=use_new_attention_order,
+            ),
+            ResBlock(
+                ch,
+                time_embed_dim,
+                dropout,
+                dims=dims,
+                use_checkpoint=use_checkpoint,
+                use_scale_shift_norm=use_scale_shift_norm,
+            ),
+        )
+        self._feature_size += ch
+        self.pool = pool
+        if pool == "adaptive":
+            self.out = nn.Sequential(
+                normalization(ch),
+                nn.SiLU(),
+                nn.AdaptiveAvgPool2d((1, 1)),
+                zero_module(conv_nd(dims, ch, out_channels, 1)),
+                nn.Flatten(),
+            )
+        elif pool == "attention":
+            assert num_head_channels != -1
+            self.out = nn.Sequential(
+                normalization(ch),
+                nn.SiLU(),
+                AttentionPool2d(
+                    (image_size // ds), ch, num_head_channels, out_channels
+                ),
+            )
+        elif pool == "spatial":
+            self.out = nn.Sequential(
+                nn.Linear(self._feature_size, 2048),
+                nn.ReLU(),
+                nn.Linear(2048, self.out_channels),
+            )
+        elif pool == "spatial_v2":
+            self.out = nn.Sequential(
+                nn.Linear(self._feature_size, 2048),
+                normalization(2048),
+                nn.SiLU(),
+                nn.Linear(2048, self.out_channels),
+            )
+        else:
+            raise NotImplementedError(f"Unexpected {pool} pooling")
+
+    def convert_to_fp16(self):
+        """
+        Convert the torso of the model to float16.
+        """
+        self.input_blocks.apply(convert_module_to_f16)
+        self.middle_block.apply(convert_module_to_f16)
+
+    def convert_to_fp32(self):
+        """
+        Convert the torso of the model to float32.
+        """
+        self.input_blocks.apply(convert_module_to_f32)
+        self.middle_block.apply(convert_module_to_f32)
+
+    def forward(self, x, timesteps):
+        """
+        Apply the model to an input batch.
+
+        :param x: an [N x C x ...] Tensor of inputs.
+        :param timesteps: a 1-D batch of timesteps.
+        :return: an [N x K] Tensor of outputs.
+        """
+        emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
+
+        results = []
+        h = x.type(self.dtype)
+        for module in self.input_blocks:
+            h = module(h, emb)
+            if self.pool.startswith("spatial"):
+                results.append(h.type(x.dtype).mean(dim=(2, 3)))
+        h = self.middle_block(h, emb)
+        if self.pool.startswith("spatial"):
+            results.append(h.type(x.dtype).mean(dim=(2, 3)))
+            h = th.cat(results, axis=-1)
+            return self.out(h)
+        else:
+            h = h.type(x.dtype)
+            return self.out(h)

guided_diffusion/model-card.md

@@ -0,0 +1,59 @@
+# Overview
+
+These are diffusion models and noised image classifiers described in the paper [Diffusion Models Beat GANs on Image Synthesis](https://arxiv.org/abs/2105.05233).
+Included in this release are the following models:
+
+ * Noisy ImageNet classifiers at resolutions 64x64, 128x128, 256x256, 512x512
+ * A class-unconditional ImageNet diffusion model at resolution 256x256
+ * Class conditional ImageNet diffusion models at 64x64, 128x128, 256x256, 512x512 resolutions
+ * Class-conditional ImageNet upsampling diffusion models: 64x64->256x256, 128x128->512x512
+ * Diffusion models trained on three LSUN classes at 256x256 resolution: cat, horse, bedroom
+
+# Datasets
+
+All of the models we are releasing were either trained on the [ILSVRC 2012 subset of ImageNet](http://www.image-net.org/challenges/LSVRC/2012/) or on single classes of [LSUN](https://arxiv.org/abs/1506.03365).
+Here, we describe characteristics of these datasets which impact model behavior:
+
+**LSUN**: This dataset was collected in 2015 using a combination of human labeling (from Amazon Mechanical Turk) and automated data labeling.
+ * Each of the three classes we consider contain over a million images.
+ * The dataset creators found that the label accuracy was roughly 90% across the entire LSUN dataset when measured by trained experts.
+ * Images are scraped from the internet, and LSUN cat images in particular tend to often follow a “meme” format.
+ * We found that there are occasionally humans in these photos, including faces, especially within the cat class.  
+
+**ILSVRC 2012 subset of ImageNet**: This dataset was curated in 2012 and consists of roughly one million images, each belonging to one of 1000 classes.
+ * A large portion of the classes in this dataset are animals, plants, and other naturally-occurring objects.
+ * Many images contain humans, although usually these humans aren’t reflected by the class label (e.g. the class “Tench, tinca tinca” contains many photos of people holding fish).
+
+# Performance
+
+These models are intended to generate samples consistent with their training distributions.
+This has been measured in terms of FID, Precision, and Recall.
+These metrics all rely on the representations of a [pre-trained Inception-V3 model](https://arxiv.org/abs/1512.00567),
+which was trained on ImageNet, and so is likely to focus more on the ImageNet classes (such as animals) than on other visual features (such as human faces).
+
+Qualitatively, the samples produced by these models often look highly realistic, especially when a diffusion model is combined with a noisy classifier.
+
+# Intended Use
+
+These models are intended to be used for research purposes only.
+In particular, they can be used as a baseline for generative modeling research, or as a starting point to build off of for such research.
+
+These models are not intended to be commercially deployed.
+Additionally, they are not intended to be used to create propaganda or offensive imagery.
+
+Before releasing these models, we probed their ability to ease the creation of targeted imagery, since doing so could be potentially harmful.
+We did this either by fine-tuning our ImageNet models on a target LSUN class, or through classifier guidance with publicly available [CLIP models](https://github.com/openai/CLIP).
+ * To probe fine-tuning capabilities, we restricted our compute budget to roughly $100 and tried both standard fine-tuning,
+and a diffusion-specific approach where we train a specialized classifier for the LSUN class. The resulting FIDs were significantly worse than publicly available GAN models, indicating that fine-tuning an ImageNet diffusion model does not significantly lower the cost of image generation.
+ * To probe guidance with CLIP, we tried two approaches for using pre-trained CLIP models for classifier guidance. Either we fed the noised image to CLIP directly and used its gradients, or we fed the diffusion model's denoised prediction to the CLIP model and differentiated through the whole process. In both cases, we found that it was difficult to recover information from the CLIP model, indicating that these diffusion models are unlikely to make it significantly easier to extract knowledge from CLIP compared to existing GAN models.
+
+# Limitations
+
+These models sometimes produce highly unrealistic outputs, particularly when generating images containing human faces.
+This may stem from ImageNet's emphasis on non-human objects.
+
+While classifier guidance can improve sample quality, it reduces diversity, resulting in some modes of the data distribution being underrepresented.
+This can potentially amplify existing biases in the training dataset such as gender and racial biases.
+
+Because ImageNet and LSUN contain images from the internet, they include photos of real people, and the model may have memorized some of the information contained in these photos.
+However, these images are already publicly available, and existing generative models trained on ImageNet have not demonstrated significant leakage of this information.

guided_diffusion/scripts/classifier_sample.py

@@ -0,0 +1,131 @@
+"""
+Like image_sample.py, but use a noisy image classifier to guide the sampling
+process towards more realistic images.
+"""
+
+import argparse
+import os
+
+import numpy as np
+import torch as th
+import torch.distributed as dist
+import torch.nn.functional as F
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.script_util import (
+    NUM_CLASSES,
+    model_and_diffusion_defaults,
+    classifier_defaults,
+    create_model_and_diffusion,
+    create_classifier,
+    add_dict_to_argparser,
+    args_to_dict,
+)
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model and diffusion...")
+    model, diffusion = create_model_and_diffusion(
+        **args_to_dict(args, model_and_diffusion_defaults().keys())
+    )
+    model.load_state_dict(
+        dist_util.load_state_dict(args.model_path, map_location="cpu")
+    )
+    model.to(dist_util.dev())
+    if args.use_fp16:
+        model.convert_to_fp16()
+    model.eval()
+
+    logger.log("loading classifier...")
+    classifier = create_classifier(**args_to_dict(args, classifier_defaults().keys()))
+    classifier.load_state_dict(
+        dist_util.load_state_dict(args.classifier_path, map_location="cpu")
+    )
+    classifier.to(dist_util.dev())
+    if args.classifier_use_fp16:
+        classifier.convert_to_fp16()
+    classifier.eval()
+
+    def cond_fn(x, t, y=None):
+        assert y is not None
+        with th.enable_grad():
+            x_in = x.detach().requires_grad_(True)
+            logits = classifier(x_in, t)
+            log_probs = F.log_softmax(logits, dim=-1)
+            selected = log_probs[range(len(logits)), y.view(-1)]
+            return th.autograd.grad(selected.sum(), x_in)[0] * args.classifier_scale
+
+    def model_fn(x, t, y=None):
+        assert y is not None
+        return model(x, t, y if args.class_cond else None)
+
+    logger.log("sampling...")
+    all_images = []
+    all_labels = []
+    while len(all_images) * args.batch_size < args.num_samples:
+        model_kwargs = {}
+        classes = th.randint(
+            low=0, high=NUM_CLASSES, size=(args.batch_size,), device=dist_util.dev()
+        )
+        model_kwargs["y"] = classes
+        sample_fn = (
+            diffusion.p_sample_loop if not args.use_ddim else diffusion.ddim_sample_loop
+        )
+        sample = sample_fn(
+            model_fn,
+            (args.batch_size, 3, args.image_size, args.image_size),
+            clip_denoised=args.clip_denoised,
+            model_kwargs=model_kwargs,
+            cond_fn=cond_fn,
+            device=dist_util.dev(),
+        )
+        sample = ((sample + 1) * 127.5).clamp(0, 255).to(th.uint8)
+        sample = sample.permute(0, 2, 3, 1)
+        sample = sample.contiguous()
+
+        gathered_samples = [th.zeros_like(sample) for _ in range(dist.get_world_size())]
+        dist.all_gather(gathered_samples, sample)  # gather not supported with NCCL
+        all_images.extend([sample.cpu().numpy() for sample in gathered_samples])
+        gathered_labels = [th.zeros_like(classes) for _ in range(dist.get_world_size())]
+        dist.all_gather(gathered_labels, classes)
+        all_labels.extend([labels.cpu().numpy() for labels in gathered_labels])
+        logger.log(f"created {len(all_images) * args.batch_size} samples")
+
+    arr = np.concatenate(all_images, axis=0)
+    arr = arr[: args.num_samples]
+    label_arr = np.concatenate(all_labels, axis=0)
+    label_arr = label_arr[: args.num_samples]
+    if dist.get_rank() == 0:
+        shape_str = "x".join([str(x) for x in arr.shape])
+        out_path = os.path.join(logger.get_dir(), f"samples_{shape_str}.npz")
+        logger.log(f"saving to {out_path}")
+        np.savez(out_path, arr, label_arr)
+
+    dist.barrier()
+    logger.log("sampling complete")
+
+
+def create_argparser():
+    defaults = dict(
+        clip_denoised=True,
+        num_samples=10000,
+        batch_size=16,
+        use_ddim=False,
+        model_path="",
+        classifier_path="",
+        classifier_scale=1.0,
+    )
+    defaults.update(model_and_diffusion_defaults())
+    defaults.update(classifier_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/classifier_train.py

@@ -0,0 +1,226 @@
+"""
+Train a noised image classifier on ImageNet.
+"""
+
+import argparse
+import os
+
+import blobfile as bf
+import torch as th
+import torch.distributed as dist
+import torch.nn.functional as F
+from torch.nn.parallel.distributed import DistributedDataParallel as DDP
+from torch.optim import AdamW
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.fp16_util import MixedPrecisionTrainer
+from guided_diffusion.image_datasets import load_data
+from guided_diffusion.resample import create_named_schedule_sampler
+from guided_diffusion.script_util import (
+    add_dict_to_argparser,
+    args_to_dict,
+    classifier_and_diffusion_defaults,
+    create_classifier_and_diffusion,
+)
+from guided_diffusion.train_util import parse_resume_step_from_filename, log_loss_dict
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model and diffusion...")
+    model, diffusion = create_classifier_and_diffusion(
+        **args_to_dict(args, classifier_and_diffusion_defaults().keys())
+    )
+    model.to(dist_util.dev())
+    if args.noised:
+        schedule_sampler = create_named_schedule_sampler(
+            args.schedule_sampler, diffusion
+        )
+
+    resume_step = 0
+    if args.resume_checkpoint:
+        resume_step = parse_resume_step_from_filename(args.resume_checkpoint)
+        if dist.get_rank() == 0:
+            logger.log(
+                f"loading model from checkpoint: {args.resume_checkpoint}... at {resume_step} step"
+            )
+            model.load_state_dict(
+                dist_util.load_state_dict(
+                    args.resume_checkpoint, map_location=dist_util.dev()
+                )
+            )
+
+    # Needed for creating correct EMAs and fp16 parameters.
+    dist_util.sync_params(model.parameters())
+
+    mp_trainer = MixedPrecisionTrainer(
+        model=model, use_fp16=args.classifier_use_fp16, initial_lg_loss_scale=16.0
+    )
+
+    model = DDP(
+        model,
+        device_ids=[dist_util.dev()],
+        output_device=dist_util.dev(),
+        broadcast_buffers=False,
+        bucket_cap_mb=128,
+        find_unused_parameters=False,
+    )
+
+    logger.log("creating data loader...")
+    data = load_data(
+        data_dir=args.data_dir,
+        batch_size=args.batch_size,
+        image_size=args.image_size,
+        class_cond=True,
+        random_crop=True,
+    )
+    if args.val_data_dir:
+        val_data = load_data(
+            data_dir=args.val_data_dir,
+            batch_size=args.batch_size,
+            image_size=args.image_size,
+            class_cond=True,
+        )
+    else:
+        val_data = None
+
+    logger.log(f"creating optimizer...")
+    opt = AdamW(mp_trainer.master_params, lr=args.lr, weight_decay=args.weight_decay)
+    if args.resume_checkpoint:
+        opt_checkpoint = bf.join(
+            bf.dirname(args.resume_checkpoint), f"opt{resume_step:06}.pt"
+        )
+        logger.log(f"loading optimizer state from checkpoint: {opt_checkpoint}")
+        opt.load_state_dict(
+            dist_util.load_state_dict(opt_checkpoint, map_location=dist_util.dev())
+        )
+
+    logger.log("training classifier model...")
+
+    def forward_backward_log(data_loader, prefix="train"):
+        batch, extra = next(data_loader)
+        labels = extra["y"].to(dist_util.dev())
+
+        batch = batch.to(dist_util.dev())
+        # Noisy images
+        if args.noised:
+            t, _ = schedule_sampler.sample(batch.shape[0], dist_util.dev())
+            batch = diffusion.q_sample(batch, t)
+        else:
+            t = th.zeros(batch.shape[0], dtype=th.long, device=dist_util.dev())
+
+        for i, (sub_batch, sub_labels, sub_t) in enumerate(
+            split_microbatches(args.microbatch, batch, labels, t)
+        ):
+            logits = model(sub_batch, timesteps=sub_t)
+            loss = F.cross_entropy(logits, sub_labels, reduction="none")
+
+            losses = {}
+            losses[f"{prefix}_loss"] = loss.detach()
+            losses[f"{prefix}_acc@1"] = compute_top_k(
+                logits, sub_labels, k=1, reduction="none"
+            )
+            losses[f"{prefix}_acc@5"] = compute_top_k(
+                logits, sub_labels, k=5, reduction="none"
+            )
+            log_loss_dict(diffusion, sub_t, losses)
+            del losses
+            loss = loss.mean()
+            if loss.requires_grad:
+                if i == 0:
+                    mp_trainer.zero_grad()
+                mp_trainer.backward(loss * len(sub_batch) / len(batch))
+
+    for step in range(args.iterations - resume_step):
+        logger.logkv("step", step + resume_step)
+        logger.logkv(
+            "samples",
+            (step + resume_step + 1) * args.batch_size * dist.get_world_size(),
+        )
+        if args.anneal_lr:
+            set_annealed_lr(opt, args.lr, (step + resume_step) / args.iterations)
+        forward_backward_log(data)
+        mp_trainer.optimize(opt)
+        if val_data is not None and not step % args.eval_interval:
+            with th.no_grad():
+                with model.no_sync():
+                    model.eval()
+                    forward_backward_log(val_data, prefix="val")
+                    model.train()
+        if not step % args.log_interval:
+            logger.dumpkvs()
+        if (
+            step
+            and dist.get_rank() == 0
+            and not (step + resume_step) % args.save_interval
+        ):
+            logger.log("saving model...")
+            save_model(mp_trainer, opt, step + resume_step)
+
+    if dist.get_rank() == 0:
+        logger.log("saving model...")
+        save_model(mp_trainer, opt, step + resume_step)
+    dist.barrier()
+
+
+def set_annealed_lr(opt, base_lr, frac_done):
+    lr = base_lr * (1 - frac_done)
+    for param_group in opt.param_groups:
+        param_group["lr"] = lr
+
+
+def save_model(mp_trainer, opt, step):
+    if dist.get_rank() == 0:
+        th.save(
+            mp_trainer.master_params_to_state_dict(mp_trainer.master_params),
+            os.path.join(logger.get_dir(), f"model{step:06d}.pt"),
+        )
+        th.save(opt.state_dict(), os.path.join(logger.get_dir(), f"opt{step:06d}.pt"))
+
+
+def compute_top_k(logits, labels, k, reduction="mean"):
+    _, top_ks = th.topk(logits, k, dim=-1)
+    if reduction == "mean":
+        return (top_ks == labels[:, None]).float().sum(dim=-1).mean().item()
+    elif reduction == "none":
+        return (top_ks == labels[:, None]).float().sum(dim=-1)
+
+
+def split_microbatches(microbatch, *args):
+    bs = len(args[0])
+    if microbatch == -1 or microbatch >= bs:
+        yield tuple(args)
+    else:
+        for i in range(0, bs, microbatch):
+            yield tuple(x[i : i + microbatch] if x is not None else None for x in args)
+
+
+def create_argparser():
+    defaults = dict(
+        data_dir="",
+        val_data_dir="",
+        noised=True,
+        iterations=150000,
+        lr=3e-4,
+        weight_decay=0.0,
+        anneal_lr=False,
+        batch_size=4,
+        microbatch=-1,
+        schedule_sampler="uniform",
+        resume_checkpoint="",
+        log_interval=10,
+        eval_interval=5,
+        save_interval=10000,
+    )
+    defaults.update(classifier_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/image_nll.py

@@ -0,0 +1,96 @@
+"""
+Approximate the bits/dimension for an image model.
+"""
+
+import argparse
+import os
+
+import numpy as np
+import torch.distributed as dist
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.image_datasets import load_data
+from guided_diffusion.script_util import (
+    model_and_diffusion_defaults,
+    create_model_and_diffusion,
+    add_dict_to_argparser,
+    args_to_dict,
+)
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model and diffusion...")
+    model, diffusion = create_model_and_diffusion(
+        **args_to_dict(args, model_and_diffusion_defaults().keys())
+    )
+    model.load_state_dict(
+        dist_util.load_state_dict(args.model_path, map_location="cpu")
+    )
+    model.to(dist_util.dev())
+    model.eval()
+
+    logger.log("creating data loader...")
+    data = load_data(
+        data_dir=args.data_dir,
+        batch_size=args.batch_size,
+        image_size=args.image_size,
+        class_cond=args.class_cond,
+        deterministic=True,
+    )
+
+    logger.log("evaluating...")
+    run_bpd_evaluation(model, diffusion, data, args.num_samples, args.clip_denoised)
+
+
+def run_bpd_evaluation(model, diffusion, data, num_samples, clip_denoised):
+    all_bpd = []
+    all_metrics = {"vb": [], "mse": [], "xstart_mse": []}
+    num_complete = 0
+    while num_complete < num_samples:
+        batch, model_kwargs = next(data)
+        batch = batch.to(dist_util.dev())
+        model_kwargs = {k: v.to(dist_util.dev()) for k, v in model_kwargs.items()}
+        minibatch_metrics = diffusion.calc_bpd_loop(
+            model, batch, clip_denoised=clip_denoised, model_kwargs=model_kwargs
+        )
+
+        for key, term_list in all_metrics.items():
+            terms = minibatch_metrics[key].mean(dim=0) / dist.get_world_size()
+            dist.all_reduce(terms)
+            term_list.append(terms.detach().cpu().numpy())
+
+        total_bpd = minibatch_metrics["total_bpd"]
+        total_bpd = total_bpd.mean() / dist.get_world_size()
+        dist.all_reduce(total_bpd)
+        all_bpd.append(total_bpd.item())
+        num_complete += dist.get_world_size() * batch.shape[0]
+
+        logger.log(f"done {num_complete} samples: bpd={np.mean(all_bpd)}")
+
+    if dist.get_rank() == 0:
+        for name, terms in all_metrics.items():
+            out_path = os.path.join(logger.get_dir(), f"{name}_terms.npz")
+            logger.log(f"saving {name} terms to {out_path}")
+            np.savez(out_path, np.mean(np.stack(terms), axis=0))
+
+    dist.barrier()
+    logger.log("evaluation complete")
+
+
+def create_argparser():
+    defaults = dict(
+        data_dir="", clip_denoised=True, num_samples=1000, batch_size=1, model_path=""
+    )
+    defaults.update(model_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/image_sample.py

@@ -0,0 +1,108 @@
+"""
+Generate a large batch of image samples from a model and save them as a large
+numpy array. This can be used to produce samples for FID evaluation.
+"""
+
+import argparse
+import os
+
+import numpy as np
+import torch as th
+import torch.distributed as dist
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.script_util import (
+    NUM_CLASSES,
+    model_and_diffusion_defaults,
+    create_model_and_diffusion,
+    add_dict_to_argparser,
+    args_to_dict,
+)
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model and diffusion...")
+    model, diffusion = create_model_and_diffusion(
+        **args_to_dict(args, model_and_diffusion_defaults().keys())
+    )
+    model.load_state_dict(
+        dist_util.load_state_dict(args.model_path, map_location="cpu")
+    )
+    model.to(dist_util.dev())
+    if args.use_fp16:
+        model.convert_to_fp16()
+    model.eval()
+
+    logger.log("sampling...")
+    all_images = []
+    all_labels = []
+    while len(all_images) * args.batch_size < args.num_samples:
+        model_kwargs = {}
+        if args.class_cond:
+            classes = th.randint(
+                low=0, high=NUM_CLASSES, size=(args.batch_size,), device=dist_util.dev()
+            )
+            model_kwargs["y"] = classes
+        sample_fn = (
+            diffusion.p_sample_loop if not args.use_ddim else diffusion.ddim_sample_loop
+        )
+        sample = sample_fn(
+            model,
+            (args.batch_size, 3, args.image_size, args.image_size),
+            clip_denoised=args.clip_denoised,
+            model_kwargs=model_kwargs,
+        )
+        sample = ((sample + 1) * 127.5).clamp(0, 255).to(th.uint8)
+        sample = sample.permute(0, 2, 3, 1)
+        sample = sample.contiguous()
+
+        gathered_samples = [th.zeros_like(sample) for _ in range(dist.get_world_size())]
+        dist.all_gather(gathered_samples, sample)  # gather not supported with NCCL
+        all_images.extend([sample.cpu().numpy() for sample in gathered_samples])
+        if args.class_cond:
+            gathered_labels = [
+                th.zeros_like(classes) for _ in range(dist.get_world_size())
+            ]
+            dist.all_gather(gathered_labels, classes)
+            all_labels.extend([labels.cpu().numpy() for labels in gathered_labels])
+        logger.log(f"created {len(all_images) * args.batch_size} samples")
+
+    arr = np.concatenate(all_images, axis=0)
+    arr = arr[: args.num_samples]
+    if args.class_cond:
+        label_arr = np.concatenate(all_labels, axis=0)
+        label_arr = label_arr[: args.num_samples]
+    if dist.get_rank() == 0:
+        shape_str = "x".join([str(x) for x in arr.shape])
+        out_path = os.path.join(logger.get_dir(), f"samples_{shape_str}.npz")
+        logger.log(f"saving to {out_path}")
+        if args.class_cond:
+            np.savez(out_path, arr, label_arr)
+        else:
+            np.savez(out_path, arr)
+
+    dist.barrier()
+    logger.log("sampling complete")
+
+
+def create_argparser():
+    defaults = dict(
+        clip_denoised=True,
+        num_samples=10000,
+        batch_size=16,
+        use_ddim=False,
+        model_path="",
+    )
+    defaults.update(model_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/image_train.py

@@ -0,0 +1,83 @@
+"""
+Train a diffusion model on images.
+"""
+
+import argparse
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.image_datasets import load_data
+from guided_diffusion.resample import create_named_schedule_sampler
+from guided_diffusion.script_util import (
+    model_and_diffusion_defaults,
+    create_model_and_diffusion,
+    args_to_dict,
+    add_dict_to_argparser,
+)
+from guided_diffusion.train_util import TrainLoop
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model and diffusion...")
+    model, diffusion = create_model_and_diffusion(
+        **args_to_dict(args, model_and_diffusion_defaults().keys())
+    )
+    model.to(dist_util.dev())
+    schedule_sampler = create_named_schedule_sampler(args.schedule_sampler, diffusion)
+
+    logger.log("creating data loader...")
+    data = load_data(
+        data_dir=args.data_dir,
+        batch_size=args.batch_size,
+        image_size=args.image_size,
+        class_cond=args.class_cond,
+    )
+
+    logger.log("training...")
+    TrainLoop(
+        model=model,
+        diffusion=diffusion,
+        data=data,
+        batch_size=args.batch_size,
+        microbatch=args.microbatch,
+        lr=args.lr,
+        ema_rate=args.ema_rate,
+        log_interval=args.log_interval,
+        save_interval=args.save_interval,
+        resume_checkpoint=args.resume_checkpoint,
+        use_fp16=args.use_fp16,
+        fp16_scale_growth=args.fp16_scale_growth,
+        schedule_sampler=schedule_sampler,
+        weight_decay=args.weight_decay,
+        lr_anneal_steps=args.lr_anneal_steps,
+    ).run_loop()
+
+
+def create_argparser():
+    defaults = dict(
+        data_dir="",
+        schedule_sampler="uniform",
+        lr=1e-4,
+        weight_decay=0.0,
+        lr_anneal_steps=0,
+        batch_size=1,
+        microbatch=-1,  # -1 disables microbatches
+        ema_rate="0.9999",  # comma-separated list of EMA values
+        log_interval=10,
+        save_interval=10000,
+        resume_checkpoint="",
+        use_fp16=False,
+        fp16_scale_growth=1e-3,
+    )
+    defaults.update(model_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/super_res_sample.py

@@ -0,0 +1,119 @@
+"""
+Generate a large batch of samples from a super resolution model, given a batch
+of samples from a regular model from image_sample.py.
+"""
+
+import argparse
+import os
+
+import blobfile as bf
+import numpy as np
+import torch as th
+import torch.distributed as dist
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.script_util import (
+    sr_model_and_diffusion_defaults,
+    sr_create_model_and_diffusion,
+    args_to_dict,
+    add_dict_to_argparser,
+)
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model...")
+    model, diffusion = sr_create_model_and_diffusion(
+        **args_to_dict(args, sr_model_and_diffusion_defaults().keys())
+    )
+    model.load_state_dict(
+        dist_util.load_state_dict(args.model_path, map_location="cpu")
+    )
+    model.to(dist_util.dev())
+    if args.use_fp16:
+        model.convert_to_fp16()
+    model.eval()
+
+    logger.log("loading data...")
+    data = load_data_for_worker(args.base_samples, args.batch_size, args.class_cond)
+
+    logger.log("creating samples...")
+    all_images = []
+    while len(all_images) * args.batch_size < args.num_samples:
+        model_kwargs = next(data)
+        model_kwargs = {k: v.to(dist_util.dev()) for k, v in model_kwargs.items()}
+        sample = diffusion.p_sample_loop(
+            model,
+            (args.batch_size, 3, args.large_size, args.large_size),
+            clip_denoised=args.clip_denoised,
+            model_kwargs=model_kwargs,
+        )
+        sample = ((sample + 1) * 127.5).clamp(0, 255).to(th.uint8)
+        sample = sample.permute(0, 2, 3, 1)
+        sample = sample.contiguous()
+
+        all_samples = [th.zeros_like(sample) for _ in range(dist.get_world_size())]
+        dist.all_gather(all_samples, sample)  # gather not supported with NCCL
+        for sample in all_samples:
+            all_images.append(sample.cpu().numpy())
+        logger.log(f"created {len(all_images) * args.batch_size} samples")
+
+    arr = np.concatenate(all_images, axis=0)
+    arr = arr[: args.num_samples]
+    if dist.get_rank() == 0:
+        shape_str = "x".join([str(x) for x in arr.shape])
+        out_path = os.path.join(logger.get_dir(), f"samples_{shape_str}.npz")
+        logger.log(f"saving to {out_path}")
+        np.savez(out_path, arr)
+
+    dist.barrier()
+    logger.log("sampling complete")
+
+
+def load_data_for_worker(base_samples, batch_size, class_cond):
+    with bf.BlobFile(base_samples, "rb") as f:
+        obj = np.load(f)
+        image_arr = obj["arr_0"]
+        if class_cond:
+            label_arr = obj["arr_1"]
+    rank = dist.get_rank()
+    num_ranks = dist.get_world_size()
+    buffer = []
+    label_buffer = []
+    while True:
+        for i in range(rank, len(image_arr), num_ranks):
+            buffer.append(image_arr[i])
+            if class_cond:
+                label_buffer.append(label_arr[i])
+            if len(buffer) == batch_size:
+                batch = th.from_numpy(np.stack(buffer)).float()
+                batch = batch / 127.5 - 1.0
+                batch = batch.permute(0, 3, 1, 2)
+                res = dict(low_res=batch)
+                if class_cond:
+                    res["y"] = th.from_numpy(np.stack(label_buffer))
+                yield res
+                buffer, label_buffer = [], []
+
+
+def create_argparser():
+    defaults = dict(
+        clip_denoised=True,
+        num_samples=10000,
+        batch_size=16,
+        use_ddim=False,
+        base_samples="",
+        model_path="",
+    )
+    defaults.update(sr_model_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/scripts/super_res_train.py

@@ -0,0 +1,98 @@
+"""
+Train a super-resolution model.
+"""
+
+import argparse
+
+import torch.nn.functional as F
+
+from guided_diffusion import dist_util, logger
+from guided_diffusion.image_datasets import load_data
+from guided_diffusion.resample import create_named_schedule_sampler
+from guided_diffusion.script_util import (
+    sr_model_and_diffusion_defaults,
+    sr_create_model_and_diffusion,
+    args_to_dict,
+    add_dict_to_argparser,
+)
+from guided_diffusion.train_util import TrainLoop
+
+
+def main():
+    args = create_argparser().parse_args()
+
+    dist_util.setup_dist()
+    logger.configure()
+
+    logger.log("creating model...")
+    model, diffusion = sr_create_model_and_diffusion(
+        **args_to_dict(args, sr_model_and_diffusion_defaults().keys())
+    )
+    model.to(dist_util.dev())
+    schedule_sampler = create_named_schedule_sampler(args.schedule_sampler, diffusion)
+
+    logger.log("creating data loader...")
+    data = load_superres_data(
+        args.data_dir,
+        args.batch_size,
+        large_size=args.large_size,
+        small_size=args.small_size,
+        class_cond=args.class_cond,
+    )
+
+    logger.log("training...")
+    TrainLoop(
+        model=model,
+        diffusion=diffusion,
+        data=data,
+        batch_size=args.batch_size,
+        microbatch=args.microbatch,
+        lr=args.lr,
+        ema_rate=args.ema_rate,
+        log_interval=args.log_interval,
+        save_interval=args.save_interval,
+        resume_checkpoint=args.resume_checkpoint,
+        use_fp16=args.use_fp16,
+        fp16_scale_growth=args.fp16_scale_growth,
+        schedule_sampler=schedule_sampler,
+        weight_decay=args.weight_decay,
+        lr_anneal_steps=args.lr_anneal_steps,
+    ).run_loop()
+
+
+def load_superres_data(data_dir, batch_size, large_size, small_size, class_cond=False):
+    data = load_data(
+        data_dir=data_dir,
+        batch_size=batch_size,
+        image_size=large_size,
+        class_cond=class_cond,
+    )
+    for large_batch, model_kwargs in data:
+        model_kwargs["low_res"] = F.interpolate(large_batch, small_size, mode="area")
+        yield large_batch, model_kwargs
+
+
+def create_argparser():
+    defaults = dict(
+        data_dir="",
+        schedule_sampler="uniform",
+        lr=1e-4,
+        weight_decay=0.0,
+        lr_anneal_steps=0,
+        batch_size=1,
+        microbatch=-1,
+        ema_rate="0.9999",
+        log_interval=10,
+        save_interval=10000,
+        resume_checkpoint="",
+        use_fp16=False,
+        fp16_scale_growth=1e-3,
+    )
+    defaults.update(sr_model_and_diffusion_defaults())
+    parser = argparse.ArgumentParser()
+    add_dict_to_argparser(parser, defaults)
+    return parser
+
+
+if __name__ == "__main__":
+    main()

guided_diffusion/setup.py

@@ -0,0 +1,7 @@
+from setuptools import setup
+
+setup(
+    name="guided-diffusion",
+    py_modules=["guided_diffusion"],
+    install_requires=["blobfile>=1.0.5", "torch", "tqdm"],
+)

main.py

@@ -0,0 +1,9 @@
+from optimization.image_editor import ImageEditor
+from optimization.arguments import get_arguments
+
+
+if __name__ == "__main__":
+    args = get_arguments()
+    image_editor = ImageEditor(args)
+    image_editor.edit_image_by_prompt()
+    # image_editor.reconstruct_image()

resizer.py

@@ -0,0 +1,198 @@
+# This code was taken from: https://github.com/assafshocher/resizer by Assaf Shocher
+import numpy as np
+import torch
+from math import pi
+from torch import nn
+
+
+class Resizer(nn.Module):
+    def __init__(self, in_shape, scale_factor=None, output_shape=None, kernel=None, antialiasing=True):
+        super(Resizer, self).__init__()
+
+        # First standardize values and fill missing arguments (if needed) by deriving scale from output shape or vice versa
+        scale_factor, output_shape = self.fix_scale_and_size(in_shape, output_shape, scale_factor)
+
+        # Choose interpolation method, each method has the matching kernel size
+        method, kernel_width = {
+            "cubic": (cubic, 4.0),
+            "lanczos2": (lanczos2, 4.0),
+            "lanczos3": (lanczos3, 6.0),
+            "box": (box, 1.0),
+            "linear": (linear, 2.0),
+            None: (cubic, 4.0)  # set default interpolation method as cubic
+        }.get(kernel)
+
+        # Antialiasing is only used when downscaling
+        antialiasing *= (np.any(np.array(scale_factor) < 1))
+
+        # Sort indices of dimensions according to scale of each dimension. since we are going dim by dim this is efficient
+        sorted_dims = np.argsort(np.array(scale_factor))
+        self.sorted_dims = [int(dim) for dim in sorted_dims if scale_factor[dim] != 1]
+
+        # Iterate over dimensions to calculate local weights for resizing and resize each time in one direction
+        field_of_view_list = []
+        weights_list = []
+        for dim in self.sorted_dims:
+            # for each coordinate (along 1 dim), calculate which coordinates in the input image affect its result and the
+            # weights that multiply the values there to get its result.
+            weights, field_of_view = self.contributions(in_shape[dim], output_shape[dim], scale_factor[dim], method,
+                                                        kernel_width, antialiasing)
+
+            # convert to torch tensor
+            weights = torch.tensor(weights.T, dtype=torch.float32)
+
+            # We add singleton dimensions to the weight matrix so we can multiply it with the big tensor we get for
+            # tmp_im[field_of_view.T], (bsxfun style)
+            weights_list.append(
+                nn.Parameter(torch.reshape(weights, list(weights.shape) + (len(scale_factor) - 1) * [1]),
+                             requires_grad=False))
+            field_of_view_list.append(
+                nn.Parameter(torch.tensor(field_of_view.T.astype(np.int32), dtype=torch.long), requires_grad=False))
+
+        self.field_of_view = nn.ParameterList(field_of_view_list)
+        self.weights = nn.ParameterList(weights_list)
+
+    def forward(self, in_tensor):
+        x = in_tensor
+
+        # Use the affecting position values and the set of weights to calculate the result of resizing along this 1 dim
+        for dim, fov, w in zip(self.sorted_dims, self.field_of_view, self.weights):
+            # To be able to act on each dim, we swap so that dim 0 is the wanted dim to resize
+            x = torch.transpose(x, dim, 0)
+
+            # This is a bit of a complicated multiplication: x[field_of_view.T] is a tensor of order image_dims+1.
+            # for each pixel in the output-image it matches the positions the influence it from the input image (along 1 dim
+            # only, this is why it only adds 1 dim to 5the shape). We then multiply, for each pixel, its set of positions with
+            # the matching set of weights. we do this by this big tensor element-wise multiplication (MATLAB bsxfun style:
+            # matching dims are multiplied element-wise while singletons mean that the matching dim is all multiplied by the
+            # same number
+            x = torch.sum(x[fov] * w, dim=0)
+
+            # Finally we swap back the axes to the original order
+            x = torch.transpose(x, dim, 0)
+
+        return x
+
+    def fix_scale_and_size(self, input_shape, output_shape, scale_factor):
+        # First fixing the scale-factor (if given) to be standardized the function expects (a list of scale factors in the
+        # same size as the number of input dimensions)
+        if scale_factor is not None:
+            # By default, if scale-factor is a scalar we assume 2d resizing and duplicate it.
+            if np.isscalar(scale_factor) and len(input_shape) > 1:
+                scale_factor = [scale_factor, scale_factor]
+
+            # We extend the size of scale-factor list to the size of the input by assigning 1 to all the unspecified scales
+            scale_factor = list(scale_factor)
+            scale_factor = [1] * (len(input_shape) - len(scale_factor)) + scale_factor
+
+        # Fixing output-shape (if given): extending it to the size of the input-shape, by assigning the original input-size
+        # to all the unspecified dimensions
+        if output_shape is not None:
+            output_shape = list(input_shape[len(output_shape):]) + list(np.uint(np.array(output_shape)))
+
+        # Dealing with the case of non-give scale-factor, calculating according to output-shape. note that this is
+        # sub-optimal, because there can be different scales to the same output-shape.
+        if scale_factor is None:
+            scale_factor = 1.0 * np.array(output_shape) / np.array(input_shape)
+
+        # Dealing with missing output-shape. calculating according to scale-factor
+        if output_shape is None:
+            output_shape = np.uint(np.ceil(np.array(input_shape) * np.array(scale_factor)))
+
+        return scale_factor, output_shape
+
+    def contributions(self, in_length, out_length, scale, kernel, kernel_width, antialiasing):
+        # This function calculates a set of 'filters' and a set of field_of_view that will later on be applied
+        # such that each position from the field_of_view will be multiplied with a matching filter from the
+        # 'weights' based on the interpolation method and the distance of the sub-pixel location from the pixel centers
+        # around it. This is only done for one dimension of the image.
+
+        # When anti-aliasing is activated (default and only for downscaling) the receptive field is stretched to size of
+        # 1/sf. this means filtering is more 'low-pass filter'.
+        fixed_kernel = (lambda arg: scale * kernel(scale * arg)) if antialiasing else kernel
+        kernel_width *= 1.0 / scale if antialiasing else 1.0
+
+        # These are the coordinates of the output image
+        out_coordinates = np.arange(1, out_length + 1)
+
+        # since both scale-factor and output size can be provided simulatneously, perserving the center of the image requires shifting
+        # the output coordinates. the deviation is because out_length doesn't necesary equal in_length*scale.
+        # to keep the center we need to subtract half of this deivation so that we get equal margins for boths sides and center is preserved.
+        shifted_out_coordinates = out_coordinates - (out_length - in_length * scale) / 2
+
+        # These are the matching positions of the output-coordinates on the input image coordinates.
+        # Best explained by example: say we have 4 horizontal pixels for HR and we downscale by SF=2 and get 2 pixels:
+        # [1,2,3,4] -> [1,2]. Remember each pixel number is the middle of the pixel.
+        # The scaling is done between the distances and not pixel numbers (the right boundary of pixel 4 is transformed to
+        # the right boundary of pixel 2. pixel 1 in the small image matches the boundary between pixels 1 and 2 in the big
+        # one and not to pixel 2. This means the position is not just multiplication of the old pos by scale-factor).
+        # So if we measure distance from the left border, middle of pixel 1 is at distance d=0.5, border between 1 and 2 is
+        # at d=1, and so on (d = p - 0.5).  we calculate (d_new = d_old / sf) which means:
+        # (p_new-0.5 = (p_old-0.5) / sf)     ->          p_new = p_old/sf + 0.5 * (1-1/sf)
+        match_coordinates = shifted_out_coordinates / scale + 0.5 * (1 - 1 / scale)
+
+        # This is the left boundary to start multiplying the filter from, it depends on the size of the filter
+        left_boundary = np.floor(match_coordinates - kernel_width / 2)
+
+        # Kernel width needs to be enlarged because when covering has sub-pixel borders, it must 'see' the pixel centers
+        # of the pixels it only covered a part from. So we add one pixel at each side to consider (weights can zeroize them)
+        expanded_kernel_width = np.ceil(kernel_width) + 2
+
+        # Determine a set of field_of_view for each each output position, these are the pixels in the input image
+        # that the pixel in the output image 'sees'. We get a matrix whos horizontal dim is the output pixels (big) and the
+        # vertical dim is the pixels it 'sees' (kernel_size + 2)
+        field_of_view = np.squeeze(
+            np.int16(np.expand_dims(left_boundary, axis=1) + np.arange(expanded_kernel_width) - 1))
+
+        # Assign weight to each pixel in the field of view. A matrix whos horizontal dim is the output pixels and the
+        # vertical dim is a list of weights matching to the pixel in the field of view (that are specified in
+        # 'field_of_view')
+        weights = fixed_kernel(1.0 * np.expand_dims(match_coordinates, axis=1) - field_of_view - 1)
+
+        # Normalize weights to sum up to 1. be careful from dividing by 0
+        sum_weights = np.sum(weights, axis=1)
+        sum_weights[sum_weights == 0] = 1.0
+        weights = 1.0 * weights / np.expand_dims(sum_weights, axis=1)
+
+        # We use this mirror structure as a trick for reflection padding at the boundaries
+        mirror = np.uint(np.concatenate((np.arange(in_length), np.arange(in_length - 1, -1, step=-1))))
+        field_of_view = mirror[np.mod(field_of_view, mirror.shape[0])]
+
+        # Get rid of  weights and pixel positions that are of zero weight
+        non_zero_out_pixels = np.nonzero(np.any(weights, axis=0))
+        weights = np.squeeze(weights[:, non_zero_out_pixels])
+        field_of_view = np.squeeze(field_of_view[:, non_zero_out_pixels])
+
+        # Final products are the relative positions and the matching weights, both are output_size X fixed_kernel_size
+        return weights, field_of_view
+
+
+# These next functions are all interpolation methods. x is the distance from the left pixel center
+
+
+def cubic(x):
+    absx = np.abs(x)
+    absx2 = absx ** 2
+    absx3 = absx ** 3
+    return ((1.5 * absx3 - 2.5 * absx2 + 1) * (absx <= 1) +
+            (-0.5 * absx3 + 2.5 * absx2 - 4 * absx + 2) * ((1 < absx) & (absx <= 2)))
+
+
+def lanczos2(x):
+    return (((np.sin(pi * x) * np.sin(pi * x / 2) + np.finfo(np.float32).eps) /
+             ((pi ** 2 * x ** 2 / 2) + np.finfo(np.float32).eps))
+            * (abs(x) < 2))
+
+
+def box(x):
+    return ((-0.5 <= x) & (x < 0.5)) * 1.0
+
+
+def lanczos3(x):
+    return (((np.sin(pi * x) * np.sin(pi * x / 3) + np.finfo(np.float32).eps) /
+             ((pi ** 2 * x ** 2 / 3) + np.finfo(np.float32).eps))
+            * (abs(x) < 3))
+
+
+def linear(x):
+    return (x + 1) * ((-1 <= x) & (x < 0)) + (1 - x) * ((0 <= x) & (x <= 1))