File size: 7,851 Bytes
5c8ef86 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 |
"""
Training script of ReferFormer
Modified from DETR (https://github.com/facebookresearch/detr)
"""
import argparse
import datetime
import json
import random
import time
from pathlib import Path
import numpy as np
import torch
from torch.utils.data import DataLoader, DistributedSampler
import util.misc as utils
import datasets.samplers as samplers
from datasets import build_dataset, get_coco_api_from_dataset
from engine import train_one_epoch, evaluate, evaluate_a2d
from models import build_model
from tools.load_pretrained_weights import pre_trained_model_to_finetune
import opts
def main(args):
args.masks = True
args.dataset_file = 'joint' # joint training of ytvos and refcoco
args.binary = 1 # only run on binary referred
utils.init_distributed_mode(args)
print("git:\n {}\n".format(utils.get_sha()))
print(args)
print(f'\n Run on {args.dataset_file} dataset.')
print('\n')
device = torch.device(args.device)
# fix the seed for reproducibility
seed = args.seed + utils.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
model, criterion, postprocessor = build_model(args)
model.to(device)
model_without_ddp = model
if args.distributed:
model = torch.nn.parallel.DistributedDataParallel(model, device_ids=[args.gpu])
model_without_ddp = model.module
# for n, p in model_without_ddp.named_parameters():
# print(n)
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('number of params:', n_parameters)
def match_name_keywords(n, name_keywords):
out = False
for b in name_keywords:
if b in n:
out = True
break
return out
param_dicts = [
{
"params":
[p for n, p in model_without_ddp.named_parameters()
if not match_name_keywords(n, args.lr_backbone_names) and not match_name_keywords(n, args.lr_text_encoder_names)
and not match_name_keywords(n, args.lr_linear_proj_names) and p.requires_grad],
"lr": args.lr,
},
{
"params": [p for n, p in model_without_ddp.named_parameters() if match_name_keywords(n, args.lr_backbone_names) and p.requires_grad],
"lr": args.lr_backbone,
},
{
"params": [p for n, p in model_without_ddp.named_parameters() if match_name_keywords(n, args.lr_text_encoder_names) and p.requires_grad],
"lr": args.lr_text_encoder,
},
{
"params": [p for n, p in model_without_ddp.named_parameters() if match_name_keywords(n, args.lr_linear_proj_names) and p.requires_grad],
"lr": args.lr * args.lr_linear_proj_mult,
}
]
optimizer = torch.optim.AdamW(param_dicts, lr=args.lr,
weight_decay=args.weight_decay)
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer, args.lr_drop)
# no validation ground truth for ytvos dataset
dataset_train = build_dataset(args.dataset_file, image_set='train', args=args)
if args.distributed:
if args.cache_mode:
sampler_train = samplers.NodeDistributedSampler(dataset_train)
else:
sampler_train = samplers.DistributedSampler(dataset_train)
else:
sampler_train = torch.utils.data.RandomSampler(dataset_train)
batch_sampler_train = torch.utils.data.BatchSampler(
sampler_train, args.batch_size, drop_last=True)
data_loader_train = DataLoader(dataset_train, batch_sampler=batch_sampler_train,
collate_fn=utils.collate_fn, num_workers=args.num_workers)
output_dir = Path(args.output_dir)
if args.resume:
if args.resume.startswith('https'):
checkpoint = torch.hub.load_state_dict_from_url(
args.resume, map_location='cpu', check_hash=True)
else:
checkpoint = torch.load(args.resume, map_location='cpu')
missing_keys, unexpected_keys = model_without_ddp.load_state_dict(checkpoint['model'], strict=False)
unexpected_keys = [k for k in unexpected_keys if not (k.endswith('total_params') or k.endswith('total_ops'))]
if len(missing_keys) > 0:
print('Missing Keys: {}'.format(missing_keys))
if len(unexpected_keys) > 0:
print('Unexpected Keys: {}'.format(unexpected_keys))
if not args.eval and 'optimizer' in checkpoint and 'lr_scheduler' in checkpoint and 'epoch' in checkpoint:
import copy
p_groups = copy.deepcopy(optimizer.param_groups)
optimizer.load_state_dict(checkpoint['optimizer'])
for pg, pg_old in zip(optimizer.param_groups, p_groups):
pg['lr'] = pg_old['lr']
pg['initial_lr'] = pg_old['initial_lr']
print(optimizer.param_groups)
lr_scheduler.load_state_dict(checkpoint['lr_scheduler'])
# todo: this is a hack for doing experiment that resume from checkpoint and also modify lr scheduler (e.g., decrease lr in advance).
args.override_resumed_lr_drop = True
if args.override_resumed_lr_drop:
print('Warning: (hack) args.override_resumed_lr_drop is set to True, so args.lr_drop would override lr_drop in resumed lr_scheduler.')
lr_scheduler.step_size = args.lr_drop
lr_scheduler.base_lrs = list(map(lambda group: group['initial_lr'], optimizer.param_groups))
lr_scheduler.step(lr_scheduler.last_epoch)
args.start_epoch = checkpoint['epoch'] + 1
if args.eval:
assert args.dataset_file == 'a2d' or args.dataset_file == 'jhmdb', \
'Only A2D-Sentences and JHMDB-Sentences datasets support evaluation'
test_stats = evaluate_a2d(model, data_loader_val, postprocessor, device, args)
return
print("Start training")
start_time = time.time()
for epoch in range(args.start_epoch, args.epochs):
if args.distributed:
sampler_train.set_epoch(epoch)
train_stats = train_one_epoch(
model, criterion, data_loader_train, optimizer, device, epoch,
args.clip_max_norm)
lr_scheduler.step()
if args.output_dir:
checkpoint_paths = [output_dir / 'checkpoint.pth']
# extra checkpoint before LR drop and every epochs
# if (epoch + 1) % args.lr_drop == 0 or (epoch + 1) % 1 == 0:
if (epoch + 1) % 1 == 0:
checkpoint_paths.append(output_dir / f'checkpoint{epoch:04}.pth')
for checkpoint_path in checkpoint_paths:
utils.save_on_master({
'model': model_without_ddp.state_dict(),
'optimizer': optimizer.state_dict(),
'lr_scheduler': lr_scheduler.state_dict(),
'epoch': epoch,
'args': args,
}, checkpoint_path)
log_stats = {**{f'train_{k}': v for k, v in train_stats.items()},
'epoch': epoch,
'n_parameters': n_parameters}
if args.output_dir and utils.is_main_process():
with (output_dir / "log.txt").open("a") as f:
f.write(json.dumps(log_stats) + "\n")
total_time = time.time() - start_time
total_time_str = str(datetime.timedelta(seconds=int(total_time)))
print('Training time {}'.format(total_time_str))
if __name__ == '__main__':
parser = argparse.ArgumentParser('ReferFormer training and evaluation script', parents=[opts.get_args_parser()])
args = parser.parse_args()
if args.output_dir:
Path(args.output_dir).mkdir(parents=True, exist_ok=True)
main(args)
|