v1

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+files/videos/B_balloon_IN.mp4 filter=lfs diff=lfs merge=lfs -text

.gitignore

@@ -0,0 +1,3 @@
+__pycache__/
+output/
+gradio_cached_examples/

README.md

@@ -4,7 +4,7 @@ emoji: 🌍
 colorFrom: pink
 colorTo: pink
 sdk: gradio
-sdk_version: 4.44.1
+sdk_version: 4.44.0
 app_file: app.py
 pinned: false
 ---

app.py

@@ -0,0 +1,240 @@
+from gradio_imageslider import ImageSlider
+import functools
+import os
+import tempfile
+import diffusers
+import gradio as gr
+import imageio as imageio
+import numpy as np
+import spaces
+import torch as torch
+from PIL import Image
+from tqdm import tqdm
+from pathlib import Path
+import gradio
+from gradio.utils import get_cache_folder
+from infer import lotus, lotus_video
+
+device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+
+def infer(path_input, seed=0):
+    name_base, name_ext = os.path.splitext(os.path.basename(path_input))
+    output_g, output_d = lotus(path_input, 'normal', seed, device)
+    if not os.path.exists("files/output"):
+        os.makedirs("files/output")
+    g_save_path = os.path.join("files/output", f"{name_base}_g{name_ext}")
+    d_save_path = os.path.join("files/output", f"{name_base}_d{name_ext}")
+    output_g.save(g_save_path)
+    output_d.save(d_save_path)
+    return [path_input, g_save_path], [path_input, d_save_path]
+
+def infer_video(path_input, seed=0):
+    frames_g, frames_d = lotus_video(path_input, 'normal', seed, device)
+    if not os.path.exists("files/output"):
+        os.makedirs("files/output")
+    name_base, _ = os.path.splitext(os.path.basename(path_input))
+    g_save_path = os.path.join("files/output", f"{name_base}_g.mp4")
+    d_save_path = os.path.join("files/output", f"{name_base}_d.mp4")
+    imageio.mimsave(g_save_path, frames_g)
+    imageio.mimsave(d_save_path, frames_d)
+    return [g_save_path, d_save_path]
+
+def run_demo_server():
+    gradio_theme = gr.themes.Default()
+
+    with gr.Blocks(
+        theme=gradio_theme,
+        title="LOTUS (Normal)",
+        css="""
+            #download {
+                height: 118px;
+            }
+            .slider .inner {
+                width: 5px;
+                background: #FFF;
+            }
+            .viewport {
+                aspect-ratio: 4/3;
+            }
+            .tabs button.selected {
+                font-size: 20px !important;
+                color: crimson !important;
+            }
+            h1 {
+                text-align: center;
+                display: block;
+            }
+            h2 {
+                text-align: center;
+                display: block;
+            }
+            h3 {
+                text-align: center;
+                display: block;
+            }
+            .md_feedback li {
+                margin-bottom: 0px !important;
+            }
+        """,
+        head="""
+            <script async src="https://www.googletagmanager.com/gtag/js?id=G-1FWSVCGZTG"></script>
+            <script>
+                window.dataLayer = window.dataLayer || [];
+                function gtag() {dataLayer.push(arguments);}
+                gtag('js', new Date());
+                gtag('config', 'G-1FWSVCGZTG');
+            </script>
+        """,
+    ) as demo:
+        gr.Markdown(
+            """
+            # LOTUS: Diffusion-based Visual Foundation Model for High-quality Dense Prediction
+            <p align="center">
+            <a title="Page" href="https://lotus3d.github.io/" target="_blank" rel="noopener noreferrer" style="display: inline-block;">
+                <img src="https://img.shields.io/badge/Project-Website-pink?logo=googlechrome&logoColor=white">
+            </a>
+            <a title="arXiv" href="https://arxiv.org/abs/2409.18124" target="_blank" rel="noopener noreferrer" style="display: inline-block;">
+                <img src="https://img.shields.io/badge/arXiv-Paper-b31b1b?logo=arxiv&logoColor=white">
+            </a>
+            <a title="Github" href="https://github.com/EnVision-Research/Lotus" target="_blank" rel="noopener noreferrer" style="display: inline-block;">
+                <img src="https://img.shields.io/github/stars/EnVision-Research/Lotus?label=GitHub%20%E2%98%85&logo=github&color=C8C" alt="badge-github-stars">
+            </a>
+            <a title="Social" href="https://x.com/haodongli00/status/1839524569058582884" target="_blank" rel="noopener noreferrer" style="display: inline-block;">
+                <img src="https://www.obukhov.ai/img/badges/badge-social.svg" alt="social">
+            </a>
+        """
+        )
+        with gr.Tabs(elem_classes=["tabs"]):
+            with gr.Tab("IMAGE"):
+                with gr.Row():
+                    with gr.Column():
+                        image_input = gr.Image(
+                            label="Input Image",
+                            type="filepath",
+                        )
+                        seed = gr.Number(
+                            label="Seed (only for Generative mode)",
+                            minimum=0,
+                            maximum=999999999,
+                        )
+                        with gr.Row():
+                            image_submit_btn = gr.Button(
+                                value="Predict Normal!", variant="primary"
+                            )
+                            image_reset_btn = gr.Button(value="Reset")
+                    with gr.Column():
+                        image_output_g = ImageSlider(
+                            label="Output (Generative)",
+                            type="filepath",
+                            interactive=False,
+                            elem_classes="slider",
+                            position=0.25,
+                        )
+                        with gr.Row():
+                            image_output_d = ImageSlider(
+                                label="Output (Discriminative)",
+                                type="filepath",
+                                interactive=False,
+                                elem_classes="slider",
+                                position=0.25,
+                            )
+
+                gr.Examples(
+                    fn=infer,
+                    examples=sorted([
+                        os.path.join("files", "images", name)
+                        for name in os.listdir(os.path.join("files", "images"))
+                    ]),
+                    inputs=[image_input],
+                    outputs=[image_output_g, image_output_d],
+                    cache_examples=True,
+                )
+
+            with gr.Tab("VIDEO"):
+                with gr.Row():
+                    with gr.Column():
+                        input_video = gr.Video(
+                            label="Input Video",
+                            autoplay=True,
+                            loop=True,
+                        )
+                        seed = gr.Number(
+                            label="Seed (only for Generative mode)",
+                            minimum=0,
+                            maximum=999999999,
+                        )
+                        with gr.Row():
+                            video_submit_btn = gr.Button(
+                                value="Predict Normal!", variant="primary"
+                            )
+                            video_reset_btn = gr.Button(value="Reset")
+                    with gr.Column():
+                        video_output_g = gr.Video(
+                            label="Output (Generative)",
+                            interactive=False,
+                            autoplay=True,
+                            loop=True,
+                            show_share_button=True,
+                        )
+                        with gr.Row():
+                            video_output_d = gr.Video(
+                                label="Output (Discriminative)",
+                                interactive=False,
+                                autoplay=True,
+                                loop=True,
+                                show_share_button=True,
+                            )
+
+                gr.Examples(
+                    fn=infer_video,
+                    examples=sorted([
+                        os.path.join("files", "videos", name)
+                        for name in os.listdir(os.path.join("files", "videos"))
+                    ]),
+                    inputs=[input_video],
+                    outputs=[video_output_g, video_output_d],
+                    cache_examples=True,
+                )
+
+        ### Image
+        image_submit_btn.click(
+            fn=infer,
+            inputs=[image_input, seed],
+            outputs=[image_output_g, image_output_d],
+        )
+        image_reset_btn.click(
+            fn=lambda: (None, None, None),
+            inputs=[],
+            outputs=[image_output_g, image_output_d],
+            queue=False,
+        )
+
+        ### Video
+        video_submit_btn.click(
+            fn=infer_video,
+            inputs=[input_video, seed],
+            outputs=[video_output_g, video_output_d],
+            queue=True,
+        )
+        video_reset_btn.click(
+            fn=lambda: (None, None, None),
+            inputs=[],
+            outputs=[video_output_g, video_output_d],
+        )
+
+        ### Server launch
+        demo.queue(
+            api_open=False,
+        ).launch(
+            server_name="0.0.0.0",
+            server_port=7860,
+        )
+
+def main():
+    os.system("pip freeze")
+    if os.path.exists("files/output"):
+        os.system("rm -rf files/output")
+    run_demo_server()
+
+if __name__ == "__main__":
+    main()

files/videos/B_balloon_IN.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f82f82460a368d3afc45dbfa87806a8b2a613d0d82baaff011c0359bf8c72f9
+size 5026185

infer.py

@@ -0,0 +1,350 @@
+# from utils.args import parse_args
+import logging
+import os
+import argparse
+from pathlib import Path
+from PIL import Image
+
+import numpy as np
+import torch
+from tqdm.auto import tqdm
+from diffusers.utils import check_min_version
+
+from pipeline import LotusGPipeline, LotusDPipeline
+from utils.image_utils import colorize_depth_map
+from utils.seed_all import seed_all
+
+from contextlib import nullcontext
+import cv2
+
+check_min_version('0.28.0.dev0')
+
+def infer_pipe(pipe, image_input, task_name, seed, device):
+    if seed is None:
+        generator = None
+    else:
+        generator = torch.Generator(device=device).manual_seed(seed)
+
+    if torch.backends.mps.is_available():
+        autocast_ctx = nullcontext()
+    else:
+        autocast_ctx = torch.autocast(pipe.device.type)
+    with autocast_ctx:
+
+        test_image = Image.open(image_input).convert('RGB')
+        test_image = np.array(test_image).astype(np.float16)
+        test_image = torch.tensor(test_image).permute(2,0,1).unsqueeze(0)
+        test_image = test_image / 127.5 - 1.0 
+        test_image = test_image.to(device)
+
+        task_emb = torch.tensor([1, 0]).float().unsqueeze(0).repeat(1, 1).to(device)
+        task_emb = torch.cat([torch.sin(task_emb), torch.cos(task_emb)], dim=-1).repeat(1, 1)
+
+        # Run
+        pred = pipe(
+            rgb_in=test_image, 
+            prompt='', 
+            num_inference_steps=1, 
+            generator=generator, 
+            # guidance_scale=0,
+            output_type='np',
+            timesteps=[999],
+            task_emb=task_emb,
+            ).images[0]
+
+        # Post-process the prediction
+        if task_name == 'depth':
+            output_npy = pred.mean(axis=-1)
+            output_color = colorize_depth_map(output_npy)
+        else:
+            output_npy = pred
+            output_color = Image.fromarray((output_npy * 255).astype(np.uint8))
+
+    return output_color
+
+def lotus_video(input_video, task_name, seed, device):
+    if task_name == 'depth':
+        model_g = 'jingheya/lotus-depth-g-v1-0'
+        model_d = 'jingheya/lotus-depth-d-v1-0'
+    else:
+        model_g = 'jingheya/lotus-normal-g-v1-0'
+        model_d = 'jingheya/lotus-normal-d-v1-0'
+
+    dtype = torch.float16
+    pipe_g = LotusGPipeline.from_pretrained(
+        model_g,
+        torch_dtype=dtype,
+    )
+    pipe_d = LotusDPipeline.from_pretrained(
+        model_d,
+        torch_dtype=dtype,
+    )
+    pipe_g.to(device)
+    pipe_d.to(device)
+    pipe_g.set_progress_bar_config(disable=True)
+    pipe_d.set_progress_bar_config(disable=True)
+    logging.info(f"Successfully loading pipeline from {model_g} and {model_d}.")
+    
+    # load the video and split it into frames
+    cap = cv2.VideoCapture(input_video)
+    frames = []
+    while True:
+        ret, frame = cap.read()
+        if not ret:
+            break
+        frames.append(frame)
+    cap.release()
+    logging.info(f"There are {len(frames)} frames in the video.")
+
+    if seed is None:
+        generator = None
+    else:
+        generator = torch.Generator(device=device).manual_seed(seed)
+
+    task_emb = torch.tensor([1, 0]).float().unsqueeze(0).repeat(1, 1).to(device)
+    task_emb = torch.cat([torch.sin(task_emb), torch.cos(task_emb)], dim=-1).repeat(1, 1)
+
+    output_g = []
+    output_d = []
+    for frame in frames:
+        if torch.backends.mps.is_available():
+            autocast_ctx = nullcontext()
+        else:
+            autocast_ctx = torch.autocast(pipe_g.device.type)
+        with autocast_ctx:
+            test_image = frame
+            test_image = np.array(test_image).astype(np.float16)
+            test_image = torch.tensor(test_image).permute(2,0,1).unsqueeze(0)
+            test_image = test_image / 127.5 - 1.0 
+            test_image = test_image.to(device)
+
+            # Run
+            pred_g = pipe_g(
+                rgb_in=test_image, 
+                prompt='', 
+                num_inference_steps=1, 
+                generator=generator, 
+                # guidance_scale=0,
+                output_type='np',
+                timesteps=[999],
+                task_emb=task_emb,
+                ).images[0]
+            pred_d = pipe_d(
+                rgb_in=test_image, 
+                prompt='', 
+                num_inference_steps=1, 
+                generator=generator, 
+                # guidance_scale=0,
+                output_type='np',
+                timesteps=[999],
+                task_emb=task_emb,
+                ).images[0]
+
+            # Post-process the prediction
+            if task_name == 'depth':
+                output_npy_g = pred_g.mean(axis=-1)
+                output_color_g = colorize_depth_map(output_npy_g)
+                output_npy_d = pred_d.mean(axis=-1)
+                output_color_d = colorize_depth_map(output_npy_d)
+            else:
+                output_npy_g = pred_g
+                output_color_g = Image.fromarray((output_npy_g * 255).astype(np.uint8))
+                output_npy_d = pred_d
+                output_color_d = Image.fromarray((output_npy_d * 255).astype(np.uint8))
+            
+            output_g.append(output_color_g)
+            output_d.append(output_color_d)
+
+    return output_g, output_d
+
+def lotus(image_input, task_name, seed, device):
+    if task_name == 'depth':
+        model_g = 'jingheya/lotus-depth-g-v1-0'
+        model_d = 'jingheya/lotus-depth-d-v1-0'
+    else:
+        model_g = 'jingheya/lotus-normal-g-v1-0'
+        model_d = 'jingheya/lotus-normal-d-v1-0'
+
+    dtype = torch.float16
+    pipe_g = LotusGPipeline.from_pretrained(
+        model_g,
+        torch_dtype=dtype,
+    )
+    pipe_d = LotusDPipeline.from_pretrained(
+        model_d,
+        torch_dtype=dtype,
+    )
+    pipe_g.to(device)
+    pipe_d.to(device)
+    pipe_g.set_progress_bar_config(disable=True)
+    pipe_d.set_progress_bar_config(disable=True)
+    logging.info(f"Successfully loading pipeline from {model_g} and {model_d}.")
+    output_g = infer_pipe(pipe_g, image_input, task_name, seed, device)
+    output_d = infer_pipe(pipe_d, image_input, task_name, seed, device)
+    return output_g, output_d
+
+def parse_args():
+    '''Set the Args'''
+    parser = argparse.ArgumentParser(
+        description="Run Lotus..."
+    )
+    # model settings
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        help="pretrained model path from hugging face or local dir",
+    )
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default="sample",
+        help="The used prediction_type. ",
+    )
+    parser.add_argument(
+        "--timestep",
+        type=int,
+        default=999,
+    )
+    parser.add_argument(
+        "--mode",
+        type=str,
+        default="regression", # "generation"
+        help="Whether to use the generation or regression pipeline."
+    )
+    parser.add_argument(
+        "--task_name",
+        type=str,
+        default="depth", # "normal"
+    )
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    
+    # inference settings
+    parser.add_argument("--seed", type=int, default=None, help="Random seed.")
+    parser.add_argument(
+        "--output_dir", type=str, required=True, help="Output directory."
+    )
+    parser.add_argument(
+        "--input_dir", type=str, required=True, help="Input directory."
+    )
+    parser.add_argument(
+        "--half_precision",
+        action="store_true",
+        help="Run with half-precision (16-bit float), might lead to suboptimal result.",
+    )
+    
+    args = parser.parse_args()
+
+    return args
+
+def main():
+    logging.basicConfig(level=logging.INFO)
+    logging.info(f"Run inference...")
+    
+    args = parse_args()
+
+    # -------------------- Preparation --------------------
+    # Random seed
+    if args.seed is not None:
+        seed_all(args.seed)
+
+    # Output directories
+    os.makedirs(args.output_dir, exist_ok=True)
+    logging.info(f"Output dir = {args.output_dir}")
+
+    output_dir_color = os.path.join(args.output_dir, f'{args.task_name}_vis')
+    output_dir_npy = os.path.join(args.output_dir, f'{args.task_name}')
+    if not os.path.exists(output_dir_color): os.makedirs(output_dir_color)
+    if not os.path.exists(output_dir_npy): os.makedirs(output_dir_npy)
+
+    # half_precision
+    if args.half_precision:
+        dtype = torch.float16
+        logging.info(f"Running with half precision ({dtype}).")
+    else:
+        dtype = torch.float16
+
+    # -------------------- Device --------------------
+    if torch.cuda.is_available():
+        device = torch.device("cuda")
+    else:
+        device = torch.device("cpu")
+        logging.warning("CUDA is not available. Running on CPU will be slow.")
+    logging.info(f"Device = {device}")
+
+    # -------------------- Data --------------------
+    root_dir = Path(args.input_dir)
+    test_images = list(root_dir.rglob('*.png')) + list(root_dir.rglob('*.jpg'))
+    test_images = sorted(test_images)
+    print('==> There are', len(test_images), 'images for validation.')
+    # -------------------- Model --------------------
+    
+    if args.mode == 'generation':
+        pipeline = LotusGPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            torch_dtype=dtype,
+        )
+    elif args.mode == 'regression':
+        pipeline = LotusDPipeline.from_pretrained(
+            args.pretrained_model_name_or_path,
+            torch_dtype=dtype,
+        )
+    else:
+        raise ValueError(f'Invalid mode: {args.mode}')
+    logging.info(f"Successfully loading pipeline from {args.pretrained_model_name_or_path}.")
+
+    pipeline = pipeline.to(device)
+    pipeline.set_progress_bar_config(disable=True)
+
+    if args.enable_xformers_memory_efficient_attention:
+        pipeline.enable_xformers_memory_efficient_attention()
+
+
+    if args.seed is None:
+        generator = None
+    else:
+        generator = torch.Generator(device=device).manual_seed(args.seed)
+
+    # -------------------- Inference and saving --------------------
+    with torch.no_grad():
+        for i in tqdm(range(len(test_images))):
+            # Preprocess validation image
+            test_image = Image.open(test_images[i]).convert('RGB')
+            test_image = np.array(test_image).astype(np.float16)
+            test_image = torch.tensor(test_image).permute(2,0,1).unsqueeze(0)
+            test_image = test_image / 127.5 - 1.0 
+            test_image = test_image.to(device)
+
+            task_emb = torch.tensor([1, 0]).float().unsqueeze(0).repeat(1, 1).to(device)
+            task_emb = torch.cat([torch.sin(task_emb), torch.cos(task_emb)], dim=-1).repeat(1, 1)
+
+            # Run
+            pred = pipeline(
+                rgb_in=test_image, 
+                prompt='', 
+                num_inference_steps=1, 
+                generator=generator, 
+                # guidance_scale=0,
+                output_type='np',
+                timesteps=[args.timestep],
+                task_emb=task_emb,
+                ).images[0]
+
+            # Post-process the prediction
+            save_file_name = os.path.basename(test_images[i])[:-4]
+            if args.task_name == 'depth':
+                output_npy = pred.mean(axis=-1)
+                output_color = colorize_depth_map(output_npy)
+            else:
+                output_npy = pred
+                output_color = Image.fromarray((output_npy * 255).astype(np.uint8))
+
+            output_color.save(os.path.join(output_dir_color, f'{save_file_name}.png'))
+            np.save(os.path.join(output_dir_npy, f'{save_file_name}.npy'), output_npy)
+    
+    print('==> Inference is done. \n==> Results saved to:', args.output_dir)
+
+if __name__ == '__main__':
+    main()

pipeline.py

@@ -0,0 +1,1284 @@
+
+import inspect
+from typing import Any, Callable, Dict, List, Optional, Union
+
+import torch
+import torch.nn.functional as F
+from packaging import version
+from transformers import CLIPImageProcessor, CLIPTextModel, CLIPTokenizer, CLIPVisionModelWithProjection
+import tensorboard
+
+from diffusers.configuration_utils import FrozenDict
+from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
+from diffusers.loaders import FromSingleFileMixin, IPAdapterMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from diffusers.models import AutoencoderKL, ImageProjection, UNet2DConditionModel
+from diffusers.models.lora import adjust_lora_scale_text_encoder
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (
+    USE_PEFT_BACKEND,
+    deprecate,
+    logging,
+    replace_example_docstring,
+    scale_lora_layers,
+    unscale_lora_layers,
+)
+from diffusers.utils.torch_utils import randn_tensor
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline, StableDiffusionMixin
+from diffusers.pipelines.stable_diffusion.pipeline_output import StableDiffusionPipelineOutput
+from diffusers.pipelines.stable_diffusion.safety_checker import StableDiffusionSafetyChecker
+from diffusers import StableDiffusionPipeline
+
+logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
+
+
+def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
+    """
+    Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
+    Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+    """
+    std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
+    std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
+    # rescale the results from guidance (fixes overexposure)
+    noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
+    # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
+    noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
+    return noise_cfg
+
+
+def retrieve_timesteps(
+    scheduler,
+    num_inference_steps: Optional[int] = None,
+    device: Optional[Union[str, torch.device]] = None,
+    timesteps: Optional[List[int]] = None,
+    **kwargs,
+):
+    """
+    Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
+    custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
+
+    Args:
+        scheduler (`SchedulerMixin`):
+            The scheduler to get timesteps from.
+        num_inference_steps (`int`):
+            The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
+            must be `None`.
+        device (`str` or `torch.device`, *optional*):
+            The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
+        timesteps (`List[int]`, *optional*):
+                Custom timesteps used to support arbitrary spacing between timesteps. If `None`, then the default
+                timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
+                must be `None`.
+
+    Returns:
+        `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
+        second element is the number of inference steps.
+    """
+    if timesteps is not None:
+        accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys())
+        if not accepts_timesteps:
+            raise ValueError(
+                f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom"
+                f" timestep schedules. Please check whether you are using the correct scheduler."
+            )
+        scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+        num_inference_steps = len(timesteps)
+    else:
+        scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
+        timesteps = scheduler.timesteps
+    return timesteps, num_inference_steps
+
+
+class DirectDiffusionPipeline(
+    DiffusionPipeline,
+    StableDiffusionMixin,
+    TextualInversionLoaderMixin,
+    LoraLoaderMixin,
+    IPAdapterMixin,
+    FromSingleFileMixin,
+):
+    r"""
+    Pipeline for text-to-image generation using Stable Diffusion.
+
+    This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods
+    implemented for all pipelines (downloading, saving, running on a particular device, etc.).
+
+    The pipeline also inherits the following loading methods:
+        - [`~loaders.TextualInversionLoaderMixin.load_textual_inversion`] for loading textual inversion embeddings
+        - [`~loaders.LoraLoaderMixin.load_lora_weights`] for loading LoRA weights
+        - [`~loaders.LoraLoaderMixin.save_lora_weights`] for saving LoRA weights
+        - [`~loaders.FromSingleFileMixin.from_single_file`] for loading `.ckpt` files
+        - [`~loaders.IPAdapterMixin.load_ip_adapter`] for loading IP Adapters
+
+    Args:
+        vae ([`AutoencoderKL`]):
+            Variational Auto-Encoder (VAE) model to encode and decode images to and from latent representations.
+        text_encoder ([`~transformers.CLIPTextModel`]):
+            Frozen text-encoder ([clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14)).
+        tokenizer ([`~transformers.CLIPTokenizer`]):
+            A `CLIPTokenizer` to tokenize text.
+        unet ([`UNet2DConditionModel`]):
+            A `UNet2DConditionModel` to denoise the encoded image latents.
+        scheduler ([`SchedulerMixin`]):
+            A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+            [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+        safety_checker ([`StableDiffusionSafetyChecker`]):
+            Classification module that estimates whether generated images could be considered offensive or harmful.
+            Please refer to the [model card](https://huggingface.co/runwayml/stable-diffusion-v1-5) for more details
+            about a model's potential harms.
+        feature_extractor ([`~transformers.CLIPImageProcessor`]):
+            A `CLIPImageProcessor` to extract features from generated images; used as inputs to the `safety_checker`.
+    """
+
+    model_cpu_offload_seq = "text_encoder->image_encoder->unet->vae"
+    _optional_components = ["safety_checker", "feature_extractor", "image_encoder"]
+    _exclude_from_cpu_offload = ["safety_checker"]
+    _callback_tensor_inputs = ["latents", "prompt_embeds", "negative_prompt_embeds"]
+
+    def __init__(
+        self,
+        vae: AutoencoderKL,
+        text_encoder: CLIPTextModel,
+        tokenizer: CLIPTokenizer,
+        unet: UNet2DConditionModel,
+        scheduler: KarrasDiffusionSchedulers,
+        safety_checker: StableDiffusionSafetyChecker,
+        feature_extractor: CLIPImageProcessor,
+        image_encoder: CLIPVisionModelWithProjection = None,
+        requires_safety_checker: bool = True,
+    ):
+        super().__init__()
+
+        if hasattr(scheduler.config, "steps_offset") and scheduler.config.steps_offset != 1:
+            deprecation_message = (
+                f"The configuration file of this scheduler: {scheduler} is outdated. `steps_offset`"
+                f" should be set to 1 instead of {scheduler.config.steps_offset}. Please make sure "
+                "to update the config accordingly as leaving `steps_offset` might led to incorrect results"
+                " in future versions. If you have downloaded this checkpoint from the Hugging Face Hub,"
+                " it would be very nice if you could open a Pull request for the `scheduler/scheduler_config.json`"
+                " file"
+            )
+            deprecate("steps_offset!=1", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(scheduler.config)
+            new_config["steps_offset"] = 1
+            scheduler._internal_dict = FrozenDict(new_config)
+
+        if hasattr(scheduler.config, "clip_sample") and scheduler.config.clip_sample is True:
+            deprecation_message = (
+                f"The configuration file of this scheduler: {scheduler} has not set the configuration `clip_sample`."
+                " `clip_sample` should be set to False in the configuration file. Please make sure to update the"
+                " config accordingly as not setting `clip_sample` in the config might lead to incorrect results in"
+                " future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it would be very"
+                " nice if you could open a Pull request for the `scheduler/scheduler_config.json` file"
+            )
+            deprecate("clip_sample not set", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(scheduler.config)
+            new_config["clip_sample"] = False
+            scheduler._internal_dict = FrozenDict(new_config)
+
+        if safety_checker is None and requires_safety_checker:
+            logger.warning(
+                f"You have disabled the safety checker for {self.__class__} by passing `safety_checker=None`. Ensure"
+                " that you abide to the conditions of the Stable Diffusion license and do not expose unfiltered"
+                " results in services or applications open to the public. Both the diffusers team and Hugging Face"
+                " strongly recommend to keep the safety filter enabled in all public facing circumstances, disabling"
+                " it only for use-cases that involve analyzing network behavior or auditing its results. For more"
+                " information, please have a look at https://github.com/huggingface/diffusers/pull/254 ."
+            )
+
+        if safety_checker is not None and feature_extractor is None:
+            raise ValueError(
+                "Make sure to define a feature extractor when loading {self.__class__} if you want to use the safety"
+                " checker. If you do not want to use the safety checker, you can pass `'safety_checker=None'` instead."
+            )
+
+        is_unet_version_less_0_9_0 = hasattr(unet.config, "_diffusers_version") and version.parse(
+            version.parse(unet.config._diffusers_version).base_version
+        ) < version.parse("0.9.0.dev0")
+        is_unet_sample_size_less_64 = hasattr(unet.config, "sample_size") and unet.config.sample_size < 64
+        if is_unet_version_less_0_9_0 and is_unet_sample_size_less_64:
+            deprecation_message = (
+                "The configuration file of the unet has set the default `sample_size` to smaller than"
+                " 64 which seems highly unlikely. If your checkpoint is a fine-tuned version of any of the"
+                " following: \n- CompVis/stable-diffusion-v1-4 \n- CompVis/stable-diffusion-v1-3 \n-"
+                " CompVis/stable-diffusion-v1-2 \n- CompVis/stable-diffusion-v1-1 \n- runwayml/stable-diffusion-v1-5"
+                " \n- runwayml/stable-diffusion-inpainting \n you should change 'sample_size' to 64 in the"
+                " configuration file. Please make sure to update the config accordingly as leaving `sample_size=32`"
+                " in the config might lead to incorrect results in future versions. If you have downloaded this"
+                " checkpoint from the Hugging Face Hub, it would be very nice if you could open a Pull request for"
+                " the `unet/config.json` file"
+            )
+            deprecate("sample_size<64", "1.0.0", deprecation_message, standard_warn=False)
+            new_config = dict(unet.config)
+            new_config["sample_size"] = 64
+            unet._internal_dict = FrozenDict(new_config)
+
+        self.register_modules(
+            vae=vae,
+            text_encoder=text_encoder,
+            tokenizer=tokenizer,
+            unet=unet,
+            scheduler=scheduler,
+            safety_checker=safety_checker,
+            feature_extractor=feature_extractor,
+            image_encoder=image_encoder,
+        )
+        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
+        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+        self.register_to_config(requires_safety_checker=requires_safety_checker)
+
+    def _encode_prompt(
+        self,
+        prompt,
+        device,
+        num_images_per_prompt,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        lora_scale: Optional[float] = None,
+        **kwargs,
+    ):
+        deprecation_message = "`_encode_prompt()` is deprecated and it will be removed in a future version. Use `encode_prompt()` instead. Also, be aware that the output format changed from a concatenated tensor to a tuple."
+        deprecate("_encode_prompt()", "1.0.0", deprecation_message, standard_warn=False)
+
+        prompt_embeds_tuple = self.encode_prompt(
+            prompt=prompt,
+            device=device,
+            num_images_per_prompt=num_images_per_prompt,
+            do_classifier_free_guidance=do_classifier_free_guidance,
+            negative_prompt=negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=lora_scale,
+            **kwargs,
+        )
+
+        # concatenate for backwards comp
+        prompt_embeds = torch.cat([prompt_embeds_tuple[1], prompt_embeds_tuple[0]])
+
+        return prompt_embeds
+
+    def encode_prompt(
+        self,
+        prompt,
+        device,
+        num_images_per_prompt,
+        do_classifier_free_guidance,
+        negative_prompt=None,
+        padding_type="do_not_pad",
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        lora_scale: Optional[float] = None,
+        clip_skip: Optional[int] = None,
+    ):
+        r"""
+        Encodes the prompt into text encoder hidden states.
+
+        Args:
+            prompt (`str` or `List[str]`, *optional*):
+                prompt to be encoded
+            device: (`torch.device`):
+                torch device
+            num_images_per_prompt (`int`):
+                number of images that should be generated per prompt
+            do_classifier_free_guidance (`bool`):
+                whether to use classifier free guidance or not
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts not to guide the image generation. If not defined, one has to pass
+                `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+                less than `1`).
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
+                provided, text embeddings will be generated from `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+                weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+                argument.
+            lora_scale (`float`, *optional*):
+                A LoRA scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+        """
+        # set lora scale so that monkey patched LoRA
+        # function of text encoder can correctly access it
+        if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+            self._lora_scale = lora_scale
+
+            # dynamically adjust the LoRA scale
+            if not USE_PEFT_BACKEND:
+                adjust_lora_scale_text_encoder(self.text_encoder, lora_scale)
+            else:
+                scale_lora_layers(self.text_encoder, lora_scale)
+
+        if prompt is not None and isinstance(prompt, str):
+            batch_size = 1
+        elif prompt is not None and isinstance(prompt, list):
+            batch_size = len(prompt)
+        else:
+            batch_size = prompt_embeds.shape[0]
+
+        if prompt_embeds is None:
+            # textual inversion: process multi-vector tokens if necessary
+            if isinstance(self, TextualInversionLoaderMixin):
+                prompt = self.maybe_convert_prompt(prompt, self.tokenizer)
+
+            text_inputs = self.tokenizer(
+                prompt,
+                padding=padding_type,
+                max_length=self.tokenizer.model_max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+            text_input_ids = text_inputs.input_ids
+            untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+
+            if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
+                text_input_ids, untruncated_ids
+            ):
+                removed_text = self.tokenizer.batch_decode(
+                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
+                )
+                logger.warning(
+                    "The following part of your input was truncated because CLIP can only handle sequences up to"
+                    f" {self.tokenizer.model_max_length} tokens: {removed_text}"
+                )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = text_inputs.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            if clip_skip is None:
+                prompt_embeds = self.text_encoder(text_input_ids.to(device), attention_mask=attention_mask)
+                prompt_embeds = prompt_embeds[0]
+            else:
+                prompt_embeds = self.text_encoder(
+                    text_input_ids.to(device), attention_mask=attention_mask, output_hidden_states=True
+                )
+                # Access the `hidden_states` first, that contains a tuple of
+                # all the hidden states from the encoder layers. Then index into
+                # the tuple to access the hidden states from the desired layer.
+                prompt_embeds = prompt_embeds[-1][-(clip_skip + 1)]
+                # We also need to apply the final LayerNorm here to not mess with the
+                # representations. The `last_hidden_states` that we typically use for
+                # obtaining the final prompt representations passes through the LayerNorm
+                # layer.
+                prompt_embeds = self.text_encoder.text_model.final_layer_norm(prompt_embeds)
+
+        if self.text_encoder is not None:
+            prompt_embeds_dtype = self.text_encoder.dtype
+        elif self.unet is not None:
+            prompt_embeds_dtype = self.unet.dtype
+        else:
+            prompt_embeds_dtype = prompt_embeds.dtype
+
+        prompt_embeds = prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+        bs_embed, seq_len, _ = prompt_embeds.shape
+        # duplicate text embeddings for each generation per prompt, using mps friendly method
+        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+        prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+        # get unconditional embeddings for classifier free guidance
+        if do_classifier_free_guidance and negative_prompt_embeds is None:
+            uncond_tokens: List[str]
+            if negative_prompt is None:
+                uncond_tokens = [""] * batch_size
+            elif prompt is not None and type(prompt) is not type(negative_prompt):
+                raise TypeError(
+                    f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+                    f" {type(prompt)}."
+                )
+            elif isinstance(negative_prompt, str):
+                uncond_tokens = [negative_prompt]
+            elif batch_size != len(negative_prompt):
+                raise ValueError(
+                    f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+                    f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+                    " the batch size of `prompt`."
+                )
+            else:
+                uncond_tokens = negative_prompt
+
+            # textual inversion: process multi-vector tokens if necessary
+            if isinstance(self, TextualInversionLoaderMixin):
+                uncond_tokens = self.maybe_convert_prompt(uncond_tokens, self.tokenizer)
+
+            max_length = prompt_embeds.shape[1]
+            uncond_input = self.tokenizer(
+                uncond_tokens,
+                padding="max_length",
+                max_length=max_length,
+                truncation=True,
+                return_tensors="pt",
+            )
+
+            if hasattr(self.text_encoder.config, "use_attention_mask") and self.text_encoder.config.use_attention_mask:
+                attention_mask = uncond_input.attention_mask.to(device)
+            else:
+                attention_mask = None
+
+            negative_prompt_embeds = self.text_encoder(
+                uncond_input.input_ids.to(device),
+                attention_mask=attention_mask,
+            )
+            negative_prompt_embeds = negative_prompt_embeds[0]
+
+        if do_classifier_free_guidance:
+            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+            seq_len = negative_prompt_embeds.shape[1]
+
+            negative_prompt_embeds = negative_prompt_embeds.to(dtype=prompt_embeds_dtype, device=device)
+
+            negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+            negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+        if isinstance(self, LoraLoaderMixin) and USE_PEFT_BACKEND:
+            # Retrieve the original scale by scaling back the LoRA layers
+            unscale_lora_layers(self.text_encoder, lora_scale)
+
+        return prompt_embeds, negative_prompt_embeds
+
+    def encode_image(self, image, device, num_images_per_prompt, output_hidden_states=None):
+        dtype = next(self.image_encoder.parameters()).dtype
+
+        if not isinstance(image, torch.Tensor):
+            image = self.feature_extractor(image, return_tensors="pt").pixel_values
+
+        image = image.to(device=device, dtype=dtype)
+        if output_hidden_states:
+            image_enc_hidden_states = self.image_encoder(image, output_hidden_states=True).hidden_states[-2]
+            image_enc_hidden_states = image_enc_hidden_states.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_enc_hidden_states = self.image_encoder(
+                torch.zeros_like(image), output_hidden_states=True
+            ).hidden_states[-2]
+            uncond_image_enc_hidden_states = uncond_image_enc_hidden_states.repeat_interleave(
+                num_images_per_prompt, dim=0
+            )
+            return image_enc_hidden_states, uncond_image_enc_hidden_states
+        else:
+            image_embeds = self.image_encoder(image).image_embeds
+            image_embeds = image_embeds.repeat_interleave(num_images_per_prompt, dim=0)
+            uncond_image_embeds = torch.zeros_like(image_embeds)
+
+            return image_embeds, uncond_image_embeds
+
+    def prepare_ip_adapter_image_embeds(
+        self, ip_adapter_image, ip_adapter_image_embeds, device, num_images_per_prompt, do_classifier_free_guidance
+    ):
+        if ip_adapter_image_embeds is None:
+            if not isinstance(ip_adapter_image, list):
+                ip_adapter_image = [ip_adapter_image]
+
+            if len(ip_adapter_image) != len(self.unet.encoder_hid_proj.image_projection_layers):
+                raise ValueError(
+                    f"`ip_adapter_image` must have same length as the number of IP Adapters. Got {len(ip_adapter_image)} images and {len(self.unet.encoder_hid_proj.image_projection_layers)} IP Adapters."
+                )
+
+            image_embeds = []
+            for single_ip_adapter_image, image_proj_layer in zip(
+                ip_adapter_image, self.unet.encoder_hid_proj.image_projection_layers
+            ):
+                output_hidden_state = not isinstance(image_proj_layer, ImageProjection)
+                single_image_embeds, single_negative_image_embeds = self.encode_image(
+                    single_ip_adapter_image, device, 1, output_hidden_state
+                )
+                single_image_embeds = torch.stack([single_image_embeds] * num_images_per_prompt, dim=0)
+                single_negative_image_embeds = torch.stack(
+                    [single_negative_image_embeds] * num_images_per_prompt, dim=0
+                )
+
+                if do_classifier_free_guidance:
+                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
+                    single_image_embeds = single_image_embeds.to(device)
+
+                image_embeds.append(single_image_embeds)
+        else:
+            repeat_dims = [1]
+            image_embeds = []
+            for single_image_embeds in ip_adapter_image_embeds:
+                if do_classifier_free_guidance:
+                    single_negative_image_embeds, single_image_embeds = single_image_embeds.chunk(2)
+                    single_image_embeds = single_image_embeds.repeat(
+                        num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
+                    )
+                    single_negative_image_embeds = single_negative_image_embeds.repeat(
+                        num_images_per_prompt, *(repeat_dims * len(single_negative_image_embeds.shape[1:]))
+                    )
+                    single_image_embeds = torch.cat([single_negative_image_embeds, single_image_embeds])
+                else:
+                    single_image_embeds = single_image_embeds.repeat(
+                        num_images_per_prompt, *(repeat_dims * len(single_image_embeds.shape[1:]))
+                    )
+                image_embeds.append(single_image_embeds)
+
+        return image_embeds
+
+    def run_safety_checker(self, image, device, dtype):
+        if self.safety_checker is None:
+            has_nsfw_concept = None
+        else:
+            if torch.is_tensor(image):
+                feature_extractor_input = self.image_processor.postprocess(image, output_type="pil")
+            else:
+                feature_extractor_input = self.image_processor.numpy_to_pil(image)
+            safety_checker_input = self.feature_extractor(feature_extractor_input, return_tensors="pt").to(device)
+            image, has_nsfw_concept = self.safety_checker(
+                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
+            )
+        return image, has_nsfw_concept
+
+    def decode_latents(self, latents):
+        deprecation_message = "The decode_latents method is deprecated and will be removed in 1.0.0. Please use VaeImageProcessor.postprocess(...) instead"
+        deprecate("decode_latents", "1.0.0", deprecation_message, standard_warn=False)
+
+        latents = 1 / self.vae.config.scaling_factor * latents
+        image = self.vae.decode(latents, return_dict=False)[0]
+        image = (image / 2 + 0.5).clamp(0, 1)
+        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloat16
+        image = image.cpu().permute(0, 2, 3, 1).float().numpy()
+        return image
+
+    def prepare_extra_step_kwargs(self, generator, eta):
+        # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+        # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+        # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+        # and should be between [0, 1]
+
+        accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        extra_step_kwargs = {}
+        if accepts_eta:
+            extra_step_kwargs["eta"] = eta
+
+        # check if the scheduler accepts generator
+        accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+        if accepts_generator:
+            extra_step_kwargs["generator"] = generator
+        return extra_step_kwargs
+
+    def check_inputs(
+        self,
+        prompt,
+        height,
+        width,
+        callback_steps,
+        negative_prompt=None,
+        prompt_embeds=None,
+        negative_prompt_embeds=None,
+        ip_adapter_image=None,
+        ip_adapter_image_embeds=None,
+        callback_on_step_end_tensor_inputs=None,
+    ):
+        if height % 8 != 0 or width % 8 != 0:
+            raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.")
+
+        if callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0):
+            raise ValueError(
+                f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+                f" {type(callback_steps)}."
+            )
+        if callback_on_step_end_tensor_inputs is not None and not all(
+            k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs
+        ):
+            raise ValueError(
+                f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}"
+            )
+
+        if prompt is not None and prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+                " only forward one of the two."
+            )
+        elif prompt is None and prompt_embeds is None:
+            raise ValueError(
+                "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+            )
+        elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+            raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+
+        if negative_prompt is not None and negative_prompt_embeds is not None:
+            raise ValueError(
+                f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+                f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+            )
+
+        if prompt_embeds is not None and negative_prompt_embeds is not None:
+            if prompt_embeds.shape != negative_prompt_embeds.shape:
+                raise ValueError(
+                    "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+                    f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+                    f" {negative_prompt_embeds.shape}."
+                )
+
+        if ip_adapter_image is not None and ip_adapter_image_embeds is not None:
+            raise ValueError(
+                "Provide either `ip_adapter_image` or `ip_adapter_image_embeds`. Cannot leave both `ip_adapter_image` and `ip_adapter_image_embeds` defined."
+            )
+
+        if ip_adapter_image_embeds is not None:
+            if not isinstance(ip_adapter_image_embeds, list):
+                raise ValueError(
+                    f"`ip_adapter_image_embeds` has to be of type `list` but is {type(ip_adapter_image_embeds)}"
+                )
+            elif ip_adapter_image_embeds[0].ndim not in [3, 4]:
+                raise ValueError(
+                    f"`ip_adapter_image_embeds` has to be a list of 3D or 4D tensors but is {ip_adapter_image_embeds[0].ndim}D"
+                )
+
+    def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
+        shape = (
+            batch_size,
+            num_channels_latents,
+            int(height) // self.vae_scale_factor,
+            int(width) // self.vae_scale_factor,
+        )
+        if isinstance(generator, list) and len(generator) != batch_size:
+            raise ValueError(
+                f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
+                f" size of {batch_size}. Make sure the batch size matches the length of the generators."
+            )
+
+        if latents is None:
+            latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
+        else:
+            latents = latents.to(device)
+
+        # scale the initial noise by the standard deviation required by the scheduler
+        latents = latents * self.scheduler.init_noise_sigma
+        return latents
+
+    # Copied from diffusers.pipelines.latent_consistency_models.pipeline_latent_consistency_text2img.LatentConsistencyModelPipeline.get_guidance_scale_embedding
+    def get_guidance_scale_embedding(
+        self, w: torch.Tensor, embedding_dim: int = 512, dtype: torch.dtype = torch.float32
+    ) -> torch.FloatTensor:
+        """
+        See https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
+
+        Args:
+            w (`torch.Tensor`):
+                Generate embedding vectors with a specified guidance scale to subsequently enrich timestep embeddings.
+            embedding_dim (`int`, *optional*, defaults to 512):
+                Dimension of the embeddings to generate.
+            dtype (`torch.dtype`, *optional*, defaults to `torch.float32`):
+                Data type of the generated embeddings.
+
+        Returns:
+            `torch.FloatTensor`: Embedding vectors with shape `(len(w), embedding_dim)`.
+        """
+        assert len(w.shape) == 1
+        w = w * 1000.0
+
+        half_dim = embedding_dim // 2
+        emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
+        emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
+        emb = w.to(dtype)[:, None] * emb[None, :]
+        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
+        if embedding_dim % 2 == 1:  # zero pad
+            emb = torch.nn.functional.pad(emb, (0, 1))
+        assert emb.shape == (w.shape[0], embedding_dim)
+        return emb
+
+    @property
+    def guidance_scale(self):
+        return self._guidance_scale
+
+    @property
+    def guidance_rescale(self):
+        return self._guidance_rescale
+
+    @property
+    def clip_skip(self):
+        return self._clip_skip
+
+    # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+    # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+    # corresponds to doing no classifier free guidance.
+    @property
+    def do_classifier_free_guidance(self):
+        return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
+
+    @property
+    def cross_attention_kwargs(self):
+        return self._cross_attention_kwargs
+
+    @property
+    def num_timesteps(self):
+        return self._num_timesteps
+
+    @property
+    def interrupt(self):
+        return self._interrupt
+
+    @torch.no_grad()
+    def __call__(
+        self,
+        rgb_in: Optional[torch.FloatTensor] = None,
+        prompt: Union[str, List[str]] = None,
+        height: Optional[int] = None,
+        width: Optional[int] = None,
+        num_inference_steps: int = 50,
+        timesteps: List[int] = None,
+        guidance_scale: float = 7.5,
+        negative_prompt: Optional[Union[str, List[str]]] = None,
+        num_images_per_prompt: Optional[int] = 1,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        negative_prompt_embeds: Optional[torch.FloatTensor] = None,
+        ip_adapter_image: Optional[PipelineImageInput] = None,
+        ip_adapter_image_embeds: Optional[List[torch.FloatTensor]] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        guidance_rescale: float = 0.0,
+        clip_skip: Optional[int] = None,
+        callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None,
+        callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+        return_intermediate_timestep_idx: Optional[int] = None,
+        **kwargs,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            height (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The height in pixels of the generated image.
+            width (`int`, *optional*, defaults to `self.unet.config.sample_size * self.vae_scale_factor`):
+                The width in pixels of the generated image.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+            guidance_scale (`float`, *optional*, defaults to 7.5):
+                A higher guidance scale value encourages the model to generate images closely linked to the text
+                `prompt` at the expense of lower image quality. Guidance scale is enabled when `guidance_scale > 1`.
+            negative_prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide what to not include in image generation. If not defined, you need to
+                pass `negative_prompt_embeds` instead. Ignored when not using guidance (`guidance_scale < 1`).
+            num_images_per_prompt (`int`, *optional*, defaults to 1):
+                The number of images to generate per prompt.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated negative text embeddings. Can be used to easily tweak text inputs (prompt weighting). If
+                not provided, `negative_prompt_embeds` are generated from the `negative_prompt` input argument.
+            ip_adapter_image: (`PipelineImageInput`, *optional*): Optional image input to work with IP Adapters.
+            ip_adapter_image_embeds (`List[torch.FloatTensor]`, *optional*):
+                Pre-generated image embeddings for IP-Adapter. It should be a list of length same as number of
+                IP-adapters. Each element should be a tensor of shape `(batch_size, num_images, emb_dim)`. It should
+                contain the negative image embedding if `do_classifier_free_guidance` is set to `True`. If not
+                provided, embeddings are computed from the `ip_adapter_image` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+            guidance_rescale (`float`, *optional*, defaults to 0.0):
+                Guidance rescale factor from [Common Diffusion Noise Schedules and Sample Steps are
+                Flawed](https://arxiv.org/pdf/2305.08891.pdf). Guidance rescale factor should fix overexposure when
+                using zero terminal SNR.
+            clip_skip (`int`, *optional*):
+                Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that
+                the output of the pre-final layer will be used for computing the prompt embeddings.
+            callback_on_step_end (`Callable`, *optional*):
+                A function that calls at the end of each denoising steps during the inference. The function is called
+                with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
+                callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
+                `callback_on_step_end_tensor_inputs`.
+            callback_on_step_end_tensor_inputs (`List`, *optional*):
+                The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
+                will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
+                `._callback_tensor_inputs` attribute of your pipeline class.
+
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+                otherwise a `tuple` is returned where the first element is a list with the generated images and the
+                second element is a list of `bool`s indicating whether the corresponding generated image contains
+                "not-safe-for-work" (nsfw) content.
+        """
+
+        callback = kwargs.pop("callback", None)
+        callback_steps = kwargs.pop("callback_steps", None)
+
+        if callback is not None:
+            deprecate(
+                "callback",
+                "1.0.0",
+                "Passing `callback` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+            )
+        if callback_steps is not None:
+            deprecate(
+                "callback_steps",
+                "1.0.0",
+                "Passing `callback_steps` as an input argument to `__call__` is deprecated, consider using `callback_on_step_end`",
+            )
+
+        # 0. Default height and width to unet
+        height, width = rgb_in.shape[2:] 
+
+        # to deal with lora scaling and other possible forward hooks
+
+        # # 1. Check inputs. Raise error if not correct
+
+        self._guidance_scale = guidance_scale
+        self._guidance_rescale = guidance_rescale
+        self._clip_skip = clip_skip
+        self._cross_attention_kwargs = cross_attention_kwargs
+        self._interrupt = False
+
+        # 2. Define call parameters
+        batch_size = rgb_in.shape[0]
+
+        device = self._execution_device
+
+        # 3. Encode input prompt
+        lora_scale = (
+            self.cross_attention_kwargs.get("scale", None) if self.cross_attention_kwargs is not None else None
+        )
+
+        prompt_embeds, negative_prompt_embeds = self.encode_prompt(
+            prompt,
+            device,
+            num_images_per_prompt,
+            self.do_classifier_free_guidance,
+            negative_prompt,
+            prompt_embeds=prompt_embeds,
+            negative_prompt_embeds=negative_prompt_embeds,
+            lora_scale=lora_scale,
+            clip_skip=self.clip_skip,
+        )
+
+        # For classifier free guidance, we need to do two forward passes.
+        # Here we concatenate the unconditional and text embeddings into a single batch
+        # to avoid doing two forward passes
+        if self.do_classifier_free_guidance:
+            prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds])
+
+        if ip_adapter_image is not None or ip_adapter_image_embeds is not None:
+            image_embeds = self.prepare_ip_adapter_image_embeds(
+                ip_adapter_image,
+                ip_adapter_image_embeds,
+                device,
+                batch_size * num_images_per_prompt,
+                self.do_classifier_free_guidance,
+            )
+
+        # 4. Prepare timesteps
+        timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
+
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels // 2
+        latents = self.prepare_latents(
+            batch_size * num_images_per_prompt,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+        
+        rgb_latents = self.vae.encode(rgb_in.to(device)).latent_dist.sample()
+        rgb_latents = rgb_latents * self.vae.config.scaling_factor
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 6.1 Add image embeds for IP-Adapter
+        added_cond_kwargs = (
+            {"image_embeds": image_embeds}
+            if (ip_adapter_image is not None or ip_adapter_image_embeds is not None)
+            else None
+        )
+
+        # 6.2 Optionally get Guidance Scale Embedding
+        timestep_cond = None
+        if self.unet.config.time_cond_proj_dim is not None:
+            guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
+            timestep_cond = self.get_guidance_scale_embedding(
+                guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
+            ).to(device=device, dtype=latents.dtype)
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                if self.interrupt:
+                    continue
+
+                # expand the latents if we are doing classifier free guidance
+                latent_model_input = torch.cat([latents] * 2) if self.do_classifier_free_guidance else latents
+                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+                
+                latent_model_input = torch.cat(
+                    [rgb_latents, latent_model_input], dim=1
+                )
+
+                # predict the noise residual
+                noise_pred = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    timestep_cond=timestep_cond,
+                    cross_attention_kwargs=self.cross_attention_kwargs,
+                    added_cond_kwargs=added_cond_kwargs,
+                    return_dict=False,
+                )[0]
+
+                # perform guidance
+                if self.do_classifier_free_guidance:
+                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+                    noise_pred = noise_pred_uncond + self.guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+                if self.do_classifier_free_guidance and self.guidance_rescale > 0.0:
+                    # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+                    noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale)
+
+                # compute the previous noisy sample x_t -> x_t-1
+                pred_latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=True)
+                latents = pred_latents[0]
+
+                if callback_on_step_end is not None:
+                    callback_kwargs = {}
+                    for k in callback_on_step_end_tensor_inputs:
+                        callback_kwargs[k] = locals()[k]
+                    callback_outputs = callback_on_step_end(self, i, t, callback_kwargs)
+
+                    latents = callback_outputs.pop("latents", latents)
+                    prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds)
+                    negative_prompt_embeds = callback_outputs.pop("negative_prompt_embeds", negative_prompt_embeds)
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+                    if callback is not None and i % callback_steps == 0:
+                        step_idx = i // getattr(self.scheduler, "order", 1)
+                        callback(step_idx, t, latents)
+
+                if return_intermediate_timestep_idx == i:
+                    latents = pred_latents[1]
+                    break
+
+        if not output_type == "latent":
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+                0
+            ]
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+        else:
+            image = latents
+            has_nsfw_concept = None
+
+        if has_nsfw_concept is None:
+            do_denormalize = [True] * image.shape[0]
+        else:
+            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
+
+class LotusDPipeline(DirectDiffusionPipeline):
+    @torch.no_grad()
+    def __call__(
+        self,
+        rgb_in: Optional[torch.FloatTensor] = None,
+        task_emb: Optional[torch.FloatTensor] = None,
+        prompt: Union[str, List[str]] = None,
+        timesteps: List[int] = None,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            rgb_input (`torch.FloatTensor`):
+                Input RGB tensor, range [-1, 1]. 
+            task_emb (`torch.FloatTensor`)
+                Task switcher for reconstruction or dense prediction (depth or normal). 
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+                otherwise a `tuple` is returned where the first element is a list with the generated images and the
+                second element is a list of `bool`s indicating whether the corresponding generated image contains
+                "not-safe-for-work" (nsfw) content.
+        """
+
+        
+       
+        # 1. Define call parameters
+        self._cross_attention_kwargs = cross_attention_kwargs
+
+        device = self._execution_device
+
+        # 2. Encode input prompt
+        prompt_embeds, _ = self.encode_prompt(
+            prompt,
+            device,
+            num_images_per_prompt=1,
+            do_classifier_free_guidance=None,
+            prompt_embeds=prompt_embeds,
+        )
+
+        # 3. Prepare timesteps
+        timesteps = torch.tensor(timesteps, device=device).long()
+
+        # 4. Prepare latent variables
+        rgb_latents = self.vae.encode(rgb_in.to(device)).latent_dist.sample()
+        rgb_latents = rgb_latents * self.vae.config.scaling_factor
+
+        # 5. Denoising
+        t = timesteps[0]
+        latent_model_input = rgb_latents
+
+        pred = self.unet(
+            latent_model_input,
+            t,
+            encoder_hidden_states=prompt_embeds,
+            cross_attention_kwargs=self.cross_attention_kwargs,
+            return_dict=False,
+            class_labels=task_emb,
+        )[0]
+
+        if not output_type == "latent":
+            image = self.vae.decode(pred / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+                0
+            ]
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+        else:
+            image = pred
+            has_nsfw_concept = None
+
+        if has_nsfw_concept is None:
+            do_denormalize = [True] * image.shape[0]
+        else:
+            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)
+
+class LotusGPipeline(DirectDiffusionPipeline):
+    @torch.no_grad()
+    def __call__(
+        self,
+        rgb_in: Optional[torch.FloatTensor] = None, # Modification 240430
+        task_emb: Optional[torch.FloatTensor] = None, 
+        prompt: Union[str, List[str]] = None,
+        num_inference_steps: int = 50,
+        timesteps: List[int] = None,
+        eta: float = 0.0,
+        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+        latents: Optional[torch.FloatTensor] = None,
+        prompt_embeds: Optional[torch.FloatTensor] = None,
+        output_type: Optional[str] = "pil",
+        return_dict: bool = True,
+        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
+        **kwargs,
+    ):
+        r"""
+        The call function to the pipeline for generation.
+
+        Args:
+            rgb_input (`torch.FloatTensor`):
+                Input RGB tensor, range [-1, 1]. 
+            task_emb (`torch.FloatTensor`)
+                The task switcher to transfer the model outout domain between prediction and reconstruction. 
+            prompt (`str` or `List[str]`, *optional*):
+                The prompt or prompts to guide image generation. If not defined, you need to pass `prompt_embeds`.
+            num_inference_steps (`int`, *optional*, defaults to 50):
+                The number of denoising steps. More denoising steps usually lead to a higher quality image at the
+                expense of slower inference.
+            timesteps (`List[int]`, *optional*):
+                Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
+                in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
+                passed will be used. Must be in descending order.
+            eta (`float`, *optional*, defaults to 0.0):
+                Corresponds to parameter eta (η) from the [DDIM](https://arxiv.org/abs/2010.02502) paper. Only applies
+                to the [`~schedulers.DDIMScheduler`], and is ignored in other schedulers.
+            generator (`torch.Generator` or `List[torch.Generator]`, *optional*):
+                A [`torch.Generator`](https://pytorch.org/docs/stable/generated/torch.Generator.html) to make
+                generation deterministic.
+            latents (`torch.FloatTensor`, *optional*):
+                Pre-generated noisy latents sampled from a Gaussian distribution, to be used as inputs for image
+                generation. Can be used to tweak the same generation with different prompts. If not provided, a latents
+                tensor is generated by sampling using the supplied random `generator`.
+            prompt_embeds (`torch.FloatTensor`, *optional*):
+                Pre-generated text embeddings. Can be used to easily tweak text inputs (prompt weighting). If not
+                provided, text embeddings are generated from the `prompt` input argument.
+            output_type (`str`, *optional*, defaults to `"pil"`):
+                The output format of the generated image. Choose between `PIL.Image` or `np.array`.
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] instead of a
+                plain tuple.
+            cross_attention_kwargs (`dict`, *optional*):
+                A kwargs dictionary that if specified is passed along to the [`AttentionProcessor`] as defined in
+                [`self.processor`](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
+        Examples:
+
+        Returns:
+            [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] or `tuple`:
+                If `return_dict` is `True`, [`~pipelines.stable_diffusion.StableDiffusionPipelineOutput`] is returned,
+                otherwise a `tuple` is returned where the first element is a list with the generated images and the
+                second element is a list of `bool`s indicating whether the corresponding generated image contains
+                "not-safe-for-work" (nsfw) content.
+        """
+
+        self._cross_attention_kwargs = cross_attention_kwargs
+
+        # 1. Default height and width to unet
+        height, width = rgb_in.shape[2:] 
+
+        # 2. Define call parameters
+        batch_size = rgb_in.shape[0]
+        device = self._execution_device
+
+        # 3. Encode input prompt
+        prompt_embeds, _ = self.encode_prompt(
+            prompt,
+            device,
+            num_images_per_prompt=1,
+            do_classifier_free_guidance=None,
+            prompt_embeds=prompt_embeds,
+        )
+
+        # 4. Prepare timesteps
+        timesteps = torch.tensor(timesteps, device=device).long()
+    
+        # 5. Prepare latent variables
+        num_channels_latents = self.unet.config.in_channels // 2
+        latents = self.prepare_latents(
+            batch_size,
+            num_channels_latents,
+            height,
+            width,
+            prompt_embeds.dtype,
+            device,
+            generator,
+            latents,
+        )
+        
+        rgb_latents = self.vae.encode(rgb_in.to(device)).latent_dist.sample()
+        rgb_latents = rgb_latents * self.vae.config.scaling_factor
+
+        # 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
+        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
+
+        # 7. Denoising loop
+        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+        self._num_timesteps = len(timesteps)
+
+        with self.progress_bar(total=num_inference_steps) as progress_bar:
+            for i, t in enumerate(timesteps):
+                latent_model_input = self.scheduler.scale_model_input(latents, t)
+                latent_model_input = torch.cat(
+                    [rgb_latents, latent_model_input], dim=1
+                )
+
+                x0_pred = self.unet(
+                    latent_model_input,
+                    t,
+                    encoder_hidden_states=prompt_embeds,
+                    cross_attention_kwargs=self.cross_attention_kwargs,
+                    return_dict=False,
+                    class_labels=task_emb,
+                )[0]
+                
+                if len(timesteps) > 1:
+                    # compute the previous noisy sample x_t -> x_t-1 
+                    latents = self.scheduler.step(x0_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                else:
+                    latents = x0_pred
+
+                # call the callback, if provided
+                if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
+                    progress_bar.update()
+
+        if not output_type == "latent":
+            image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False, generator=generator)[
+                0
+            ]
+            image, has_nsfw_concept = self.run_safety_checker(image, device, prompt_embeds.dtype)
+        else:
+            image = latents
+            has_nsfw_concept = None
+
+        if has_nsfw_concept is None:
+            do_denormalize = [True] * image.shape[0]
+        else:
+            do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]
+
+        image = self.image_processor.postprocess(image, output_type=output_type, do_denormalize=do_denormalize)
+
+        # Offload all models
+        self.maybe_free_model_hooks()
+
+        if not return_dict:
+            return (image, has_nsfw_concept)
+
+        return StableDiffusionPipelineOutput(images=image, nsfw_content_detected=has_nsfw_concept)

requirements.txt

@@ -0,0 +1,25 @@
+numpy==1.26.4
+torch==2.3.1 --index-url https://download.pytorch.org/whl/cu121
+torchvision==0.18.1 --index-url https://download.pytorch.org/whl/cu121
+torchaudio==2.3.1 --index-url https://download.pytorch.org/whl/cu121
+diffusers==0.28.0
+accelerate>=0.16.0
+transformers>=4.25.1
+datasets==2.21.0
+ftfy==6.2.3
+tensorboard==2.17.1
+Jinja2==3.1.3
+peft==0.7.0
+bitsandbytes==0.44.1
+geffnet==1.0.2
+opencv-python==4.10.0.82
+matplotlib==3.8.4
+h5py==3.11.0
+omegaconf==2.3.0
+tabulate==0.9.0
+imageio==2.35.1
+imageio-ffmpeg==0.5.1
+spaces==0.28.3
+gradio==4.44.0
+gradio-imageslider==0.0.16
+gradio-client==1.3.0

utils/args.py

@@ -0,0 +1,390 @@
+import os
+import argparse
+
+def parse_args():
+    parser = argparse.ArgumentParser(description="Simple example of a training script.")
+    parser.add_argument(
+        "--input_perturbation", type=float, default=0, help="The scale of input perturbation. Recommended 0.1."
+    )
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        required=True,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--revision",
+        type=str,
+        default=None,
+        required=False,
+        help="Revision of pretrained model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--variant",
+        type=str,
+        default=None,
+        help="Variant of the model files of the pretrained model identifier from huggingface.co/models, 'e.g.' fp16",
+    )
+    parser.add_argument(
+        "--dataset_name",
+        type=str,
+        default=None,
+        help=(
+            "The name of the Dataset (from the HuggingFace hub) to train on (could be your own, possibly private,"
+            " dataset). It can also be a path pointing to a local copy of a dataset in your filesystem,"
+            " or to a folder containing files that 🤗 Datasets can understand."
+        ),
+    )
+    parser.add_argument(
+        "--dataset_config_name",
+        type=str,
+        default=None,
+        help="The config of the Dataset, leave as None if there's only one config.",
+    )
+    parser.add_argument(
+        "--train_data_dir_hypersim",
+        type=str,
+        default=None,
+        help=(
+            "A folder containing the training data. Folder contents must follow the structure described in"
+            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
+            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
+        ),
+    )
+    parser.add_argument(
+        "--train_data_dir_vkitti",
+        type=str,
+        default=None,
+        help=(
+            "A folder containing the training data. Folder contents must follow the structure described in"
+            " https://huggingface.co/docs/datasets/image_dataset#imagefolder. In particular, a `metadata.jsonl` file"
+            " must exist to provide the captions for the images. Ignored if `dataset_name` is specified."
+        ),
+    )
+    parser.add_argument(
+        "--image_column", type=str, default="image", help="The column of the dataset containing an image."
+    )
+    parser.add_argument(
+        "--depth_column", type=str, default="depth", help="The column of the dataset containing a depth file."
+    )
+    parser.add_argument(
+        "--caption_column",
+        type=str,
+        default="text",
+        help="The column of the dataset containing a caption or a list of captions.",
+    )
+    parser.add_argument(
+        "--max_train_samples",
+        type=int,
+        default=None,
+        help=(
+            "For debugging purposes or quicker training, truncate the number of training examples to this "
+            "value if set."
+        ),
+    )
+    parser.add_argument(
+        "--timestep",
+        type=int,
+        default=999,
+    )
+    parser.add_argument(
+        "--base_test_data_dir",
+        type=str,
+        default="datasets/eval/"
+    )
+    parser.add_argument(
+        "--task_name",
+        type=str,
+        default="depth", # "normal"
+    )
+    parser.add_argument(
+        "--validation_images",
+        type=str,
+        default=None,
+        help=("A set of images evaluated every `--validation_steps` and logged to `--report_to`."),
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="sd-model-finetuned",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--cache_dir",
+        type=str,
+        default=None,
+        help="The directory where the downloaded models and datasets will be stored.",
+    )
+    parser.add_argument("--seed", type=int, default=None, help="A seed for reproducible training.")
+    parser.add_argument(
+        "--resolution_hypersim",
+        type=int,
+        default=512,
+        help=(
+            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
+            " resolution"
+        ),
+    )
+    parser.add_argument(
+        "--resolution_vkitti",
+        type=int,
+        default=512,
+        help=(
+            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
+            " resolution"
+        ),
+    )
+    parser.add_argument(
+        "--prob_hypersim",
+        type=float,
+        default=0.9,
+    )
+    parser.add_argument(
+        "--mix_dataset",
+        action="store_true"
+    )
+    parser.add_argument(
+        "--mode",
+        type=str,
+        default="regression", # "generation"
+        help="Whether to use the generation or regression pipeline."
+    )
+    parser.add_argument(
+        "--norm_type",
+        type=str,
+        choices=['instnorm','truncnorm'],
+        default='truncnorm'
+    )
+    parser.add_argument(
+        "--center_crop",
+        default=False,
+        action="store_true",
+        help=(
+            "Whether to center crop the input images to the resolution. If not set, the images will be randomly"
+            " cropped. The images will be resized to the resolution first before cropping."
+        ),
+    )
+    parser.add_argument(
+        "--random_flip",
+        action="store_true",
+        help="whether to randomly flip images horizontally",
+    )
+    parser.add_argument(
+        "--train_batch_size", type=int, default=16, help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument("--num_train_epochs", type=int, default=100)
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate",
+        type=float,
+        default=1e-4,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        action="store_true",
+        default=False,
+        help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
+    )
+    parser.add_argument(
+        "--lr_scheduler",
+        type=str,
+        default="constant",
+        help=(
+            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
+            ' "constant", "constant_with_warmup"]'
+        ),
+    )
+    parser.add_argument(
+        "--lr_warmup_steps", type=int, default=500, help="Number of steps for the warmup in the lr scheduler."
+    )
+    parser.add_argument(
+        "--snr_gamma",
+        type=float,
+        default=None,
+        help="SNR weighting gamma to be used if rebalancing the loss. Recommended value is 5.0. "
+        "More details here: https://arxiv.org/abs/2303.09556.",
+    )
+    parser.add_argument(
+        "--dream_training",
+        action="store_true",
+        help=(
+            "Use the DREAM training method, which makes training more efficient and accurate at the ",
+            "expense of doing an extra forward pass. See: https://arxiv.org/abs/2312.00210",
+        ),
+    )
+    parser.add_argument(
+        "--dream_detail_preservation",
+        type=float,
+        default=1.0,
+        help="Dream detail preservation factor p (should be greater than 0; default=1.0, as suggested in the paper)",
+    )
+    parser.add_argument(
+        "--use_8bit_adam", action="store_true", help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    parser.add_argument(
+        "--allow_tf32",
+        action="store_true",
+        help=(
+            "Whether or not to allow TF32 on Ampere GPUs. Can be used to speed up training. For more information, see"
+            " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
+        ),
+    )
+    parser.add_argument("--use_ema", action="store_true", help="Whether to use EMA model.")
+    parser.add_argument(
+        "--non_ema_revision",
+        type=str,
+        default=None,
+        required=False,
+        help=(
+            "Revision of pretrained non-ema model identifier. Must be a branch, tag or git identifier of the local or"
+            " remote repository specified with --pretrained_model_name_or_path."
+        ),
+    )
+    parser.add_argument(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "Number of subprocesses to use for data loading. 0 means that the data will be loaded in the main process."
+        ),
+    )
+    parser.add_argument("--adam_beta1", type=float, default=0.9, help="The beta1 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_beta2", type=float, default=0.999, help="The beta2 parameter for the Adam optimizer.")
+    parser.add_argument("--adam_weight_decay", type=float, default=1e-2, help="Weight decay to use.")
+    parser.add_argument("--adam_epsilon", type=float, default=1e-08, help="Epsilon value for the Adam optimizer")
+    parser.add_argument("--max_grad_norm", default=1.0, type=float, help="Max gradient norm.")
+    parser.add_argument("--push_to_hub", action="store_true", help="Whether or not to push the model to the Hub.")
+    parser.add_argument("--hub_token", type=str, default=None, help="The token to use to push to the Model Hub.")
+    parser.add_argument(
+        "--prediction_type",
+        type=str,
+        default="sample",
+        help="The used prediction_type. ",
+    )
+    parser.add_argument(
+        "--hub_model_id",
+        type=str,
+        default=None,
+        help="The name of the repository to keep in sync with the local `output_dir`.",
+    )
+    parser.add_argument(
+        "--logging_dir",
+        type=str,
+        default="logs",
+        help=(
+            "[TensorBoard](https://www.tensorflow.org/tensorboard) log directory. Will default to"
+            " *output_dir/runs/**CURRENT_DATETIME_HOSTNAME***."
+        ),
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default=None,
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >="
+            " 1.10.and an Nvidia Ampere GPU.  Default to the value of accelerate config of the current system or the"
+            " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
+        ),
+    )
+    parser.add_argument(
+        "--report_to",
+        type=str,
+        default="tensorboard",
+        help=(
+            'The integration to report the results and logs to. Supported platforms are `"tensorboard"`'
+            ' (default), `"wandb"` and `"comet_ml"`. Use `"all"` to report to all integrations.'
+        ),
+    )
+    parser.add_argument("--local_rank", type=int, default=-1, help="For distributed training: local_rank")
+    parser.add_argument(
+        "--checkpointing_steps",
+        type=int,
+        default=500,
+        help=(
+            "Save a checkpoint of the training state every X updates. These checkpoints are only suitable for resuming"
+            " training using `--resume_from_checkpoint`."
+        ),
+    )
+    parser.add_argument(
+        "--checkpoints_total_limit",
+        type=int,
+        default=None,
+        help=("Max number of checkpoints to store."),
+    )
+    parser.add_argument(
+        "--resume_from_checkpoint",
+        type=str,
+        default=None,
+        help=(
+            "Whether training should be resumed from a previous checkpoint. Use a path saved by"
+            ' `--checkpointing_steps`, or `"latest"` to automatically select the last available checkpoint.'
+        ),
+    )
+    parser.add_argument(
+        "--checkpoint_dir",
+        type=str,
+        default=None,
+    )
+    parser.add_argument(
+        "--enable_xformers_memory_efficient_attention", action="store_true", help="Whether or not to use xformers."
+    )
+    parser.add_argument("--noise_offset", type=float, default=0, help="The scale of noise offset.")
+    parser.add_argument("--use_pretrained_sd", action="store_true")
+    parser.add_argument(
+        "--truncnorm_min",
+        type=float,
+        default=0.02,
+    )
+    parser.add_argument(
+        "--validation_steps",
+        type=int,
+        default=500,
+        help="Run validation every X steps.",
+    )
+    parser.add_argument(
+        "--tracker_project_name",
+        type=str,
+        default="text2image-fine-tune",
+        help=(
+            "The `project_name` argument passed to Accelerator.init_trackers for"
+            " more information see https://huggingface.co/docs/accelerate/v0.17.0/en/package_reference/accelerator#accelerate.Accelerator"
+        ),
+    )
+    parser.add_argument(
+        "--inference",
+        action="store_true"
+    )
+
+    args = parser.parse_args()
+    env_local_rank = int(os.environ.get("LOCAL_RANK", -1))
+    if env_local_rank != -1 and env_local_rank != args.local_rank:
+        args.local_rank = env_local_rank
+
+    # Sanity checks
+    if not args.inference and args.dataset_name is None and args.train_data_dir_hypersim is None:
+        raise ValueError("Need either a dataset name or a training folder.")
+
+    # default to using the same revision for the non-ema model if not specified
+    if args.non_ema_revision is None:
+        args.non_ema_revision = args.revision
+
+    return args

utils/image_utils.py

@@ -0,0 +1,107 @@
+from PIL import Image
+import matplotlib
+import numpy as np
+
+from PIL import Image
+
+import torch
+from torchvision.transforms import InterpolationMode
+from torchvision.transforms.functional import resize
+
+def concatenate_images(*image_lists):
+    # Ensure at least one image list is provided
+    if not image_lists or not image_lists[0]:
+        raise ValueError("At least one non-empty image list must be provided")
+    
+    # Determine the maximum width of any single row and the total height
+    max_width = 0
+    total_height = 0
+    row_widths = []
+    row_heights = []
+
+    # Compute dimensions for each row
+    for image_list in image_lists:
+        if image_list:  # Ensure the list is not empty
+            width = sum(img.width for img in image_list)
+            height = image_list[0].height  # Assuming all images in the list have the same height
+            max_width = max(max_width, width)
+            total_height += height
+            row_widths.append(width)
+            row_heights.append(height)
+    
+    # Create a new image to concatenate everything into
+    new_image = Image.new('RGB', (max_width, total_height))
+    
+    # Concatenate each row of images
+    y_offset = 0
+    for i, image_list in enumerate(image_lists):
+        x_offset = 0
+        for img in image_list:
+            new_image.paste(img, (x_offset, y_offset))
+            x_offset += img.width
+        y_offset += row_heights[i]  # Move the offset down to the next row
+    
+    return new_image
+
+
+def colorize_depth_map(depth, mask=None):
+    cm = matplotlib.colormaps["Spectral"]
+    # normalize
+    depth = ((depth - depth.min()) / (depth.max() - depth.min()))
+    # colorize
+    img_colored_np = cm(depth, bytes=False)[:, :, 0:3] # (h,w,3)
+    depth_colored = (img_colored_np * 255).astype(np.uint8) 
+    if mask is not None:
+        masked_image = np.zeros_like(depth_colored)
+        masked_image[mask.numpy()] = depth_colored[mask.numpy()]
+        depth_colored_img = Image.fromarray(masked_image)
+    else:
+        depth_colored_img = Image.fromarray(depth_colored)
+    return depth_colored_img
+
+
+def resize_max_res(
+    img: torch.Tensor,
+    max_edge_resolution: int,
+    resample_method: InterpolationMode = InterpolationMode.BILINEAR,
+) -> torch.Tensor:
+    """
+    Resize image to limit maximum edge length while keeping aspect ratio.
+
+    Args:
+        img (`torch.Tensor`):
+            Image tensor to be resized. Expected shape: [B, C, H, W]
+        max_edge_resolution (`int`):
+            Maximum edge length (pixel).
+        resample_method (`PIL.Image.Resampling`):
+            Resampling method used to resize images.
+
+    Returns:
+        `torch.Tensor`: Resized image.
+    """
+    assert 4 == img.dim(), f"Invalid input shape {img.shape}"
+
+    original_height, original_width = img.shape[-2:]
+    downscale_factor = min(
+        max_edge_resolution / original_width, max_edge_resolution / original_height
+    )
+
+    new_width = int(original_width * downscale_factor)
+    new_height = int(original_height * downscale_factor)
+
+    resized_img = resize(img, (new_height, new_width), resample_method, antialias=True)
+    return resized_img
+
+
+def get_tv_resample_method(method_str: str) -> InterpolationMode:
+    resample_method_dict = {
+        "bilinear": InterpolationMode.BILINEAR,
+        "bicubic": InterpolationMode.BICUBIC,
+        "nearest": InterpolationMode.NEAREST_EXACT,
+        "nearest-exact": InterpolationMode.NEAREST_EXACT,
+    }
+    resample_method = resample_method_dict.get(method_str, None)
+    if resample_method is None:
+        raise ValueError(f"Unknown resampling method: {resample_method}")
+    else:
+        return resample_method

utils/seed_all.py

@@ -0,0 +1,33 @@
+# Copyright 2023 Bingxin Ke, ETH Zurich. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# --------------------------------------------------------------------------
+# If you find this code useful, we kindly ask you to cite our paper in your work.
+# Please find bibtex at: https://github.com/prs-eth/Marigold#-citation
+# More information about the method can be found at https://marigoldmonodepth.github.io
+# --------------------------------------------------------------------------
+
+
+import numpy as np
+import random
+import torch
+
+
+def seed_all(seed: int = 0):
+    """
+    Set random seeds of all components.
+    """
+    random.seed(seed)
+    np.random.seed(seed)
+    torch.manual_seed(seed)
+    torch.cuda.manual_seed_all(seed)

utils/visualize.py

@@ -0,0 +1,119 @@
+import cv2
+import numpy as np
+
+import torch
+
+from matplotlib import cm
+import matplotlib.pyplot as plt
+
+import logging
+logger = logging.getLogger('root')
+
+
+
+def tensor_to_numpy(tensor_in):
+    """ torch tensor to numpy array
+    """
+    if tensor_in is not None:
+        if tensor_in.ndim == 3:
+            # (C, H, W) -> (H, W, C)
+            tensor_in = tensor_in.detach().cpu().permute(1, 2, 0).numpy()
+        elif tensor_in.ndim == 4:
+            # (B, C, H, W) -> (B, H, W, C)
+            tensor_in = tensor_in.detach().cpu().permute(0, 2, 3, 1).numpy()
+        else:
+            raise Exception('invalid tensor size')
+    return tensor_in
+
+# def unnormalize(img_in, img_stats={'mean': [0.485, 0.456, 0.406], 
+#                                     'std': [0.229, 0.224, 0.225]}):
+def unnormalize(img_in, img_stats={'mean': [0.5,0.5,0.5], 'std': [0.5,0.5,0.5]}):
+    """ unnormalize input image
+    """
+    if torch.is_tensor(img_in):
+        img_in = tensor_to_numpy(img_in)
+
+    img_out = np.zeros_like(img_in)
+    for ich in range(3):
+        img_out[..., ich] = img_in[..., ich] * img_stats['std'][ich]
+        img_out[..., ich] += img_stats['mean'][ich]
+    img_out = (img_out * 255.0).astype(np.uint8)
+    return img_out
+
+def normal_to_rgb(normal, normal_mask=None):
+    """ surface normal map to RGB
+        (used for visualization)
+
+        NOTE: x, y, z are mapped to R, G, B
+        NOTE: [-1, 1] are mapped to [0, 255]
+    """
+    if torch.is_tensor(normal):
+        normal = tensor_to_numpy(normal)
+        normal_mask = tensor_to_numpy(normal_mask)
+
+    normal_norm = np.linalg.norm(normal, axis=-1, keepdims=True)
+    normal_norm[normal_norm < 1e-12] = 1e-12
+    normal = normal / normal_norm
+
+    normal_rgb = (((normal + 1) * 0.5) * 255).astype(np.uint8)
+    if normal_mask is not None:
+        normal_rgb = normal_rgb * normal_mask     # (B, H, W, 3)
+    return normal_rgb
+
+def kappa_to_alpha(pred_kappa, to_numpy=True):
+    """ Confidence kappa to uncertainty alpha
+        Assuming AngMF distribution (introduced in https://arxiv.org/abs/2109.09881)
+    """
+    if torch.is_tensor(pred_kappa) and to_numpy:
+        pred_kappa = tensor_to_numpy(pred_kappa)
+
+    if torch.is_tensor(pred_kappa):
+        alpha = ((2 * pred_kappa) / ((pred_kappa ** 2.0) + 1)) \
+            + ((torch.exp(- pred_kappa * np.pi) * np.pi) / (1 + torch.exp(- pred_kappa * np.pi)))
+        alpha = torch.rad2deg(alpha)
+    else:
+        alpha = ((2 * pred_kappa) / ((pred_kappa ** 2.0) + 1)) \
+                + ((np.exp(- pred_kappa * np.pi) * np.pi) / (1 + np.exp(- pred_kappa * np.pi)))
+        alpha = np.degrees(alpha)
+
+    return alpha
+
+
+def visualize_normal(target_dir, prefixs, img, pred_norm, pred_kappa,
+                        gt_norm, gt_norm_mask, pred_error, num_vis=-1):
+    """ visualize normal
+    """
+    error_max = 60.0
+
+    img = tensor_to_numpy(img)                      # (B, H, W, 3)
+    pred_norm = tensor_to_numpy(pred_norm)          # (B, H, W, 3)
+    pred_kappa = tensor_to_numpy(pred_kappa)        # (B, H, W, 1)
+    gt_norm = tensor_to_numpy(gt_norm)              # (B, H, W, 3)
+    gt_norm_mask = tensor_to_numpy(gt_norm_mask)    # (B, H, W, 1)
+    pred_error = tensor_to_numpy(pred_error)        # (B, H, W, 1)
+
+    num_vis = len(prefixs) if num_vis == -1 else num_vis
+    for i in range(num_vis):
+        # img
+        img_ = unnormalize(img[i, ...])
+        target_path = '%s/%s_img.png' % (target_dir, prefixs[i])
+        plt.imsave(target_path, img_)
+
+        # pred_norm 
+        target_path = '%s/%s_norm.png' % (target_dir, prefixs[i])
+        plt.imsave(target_path, normal_to_rgb(pred_norm[i, ...]))
+
+        # pred_kappa
+        if pred_kappa is not None:
+            pred_alpha = kappa_to_alpha(pred_kappa[i, :, :, 0])
+            target_path = '%s/%s_pred_alpha.png' % (target_dir, prefixs[i])
+            plt.imsave(target_path, pred_alpha, vmin=0.0, vmax=error_max, cmap='jet')
+
+        # gt_norm, pred_error
+        if gt_norm is not None:
+            target_path = '%s/%s_gt.png' % (target_dir, prefixs[i])
+            plt.imsave(target_path, normal_to_rgb(gt_norm[i, ...], gt_norm_mask[i, ...]))
+
+            E = pred_error[i, :, :, 0] * gt_norm_mask[i, :, :, 0]
+            target_path = '%s/%s_pred_error.png' % (target_dir, prefixs[i])
+            plt.imsave(target_path, E, vmin=0, vmax=error_max, cmap='jet')