rm tune-a-video, add default config for video crop

Dockerfile

@@ -53,7 +53,6 @@ COPY --chown=1000 . ${HOME}/app
 RUN ls -a
 RUN cd ./FateZero/ckpt && bash download.sh
 RUN cd ./FateZero/data && bash download.sh
-RUN cd Tune-A-Video && patch -p1 < ../patch
 ENV PYTHONPATH=${HOME}/app \
     PYTHONUNBUFFERED=1 \
     GRADIO_ALLOW_FLAGGING=never \

FateZero/test_fatezero.py

@@ -260,7 +260,7 @@ def run(config='FateZero/config/low_resource_teaser/jeep_watercolor_ddim_10_step
     if 'unet' in os.listdir(Omegadict['pretrained_model_path']):
         test(config=config, **Omegadict)
         print('test finished')
-        return '/home/cqiaa/diffusion/hugging_face/Tune-A-Video-inference/FateZero/result/low_resource_teaser/jeep_watercolor_ddim_10_steps_230327-200651/sample/step_0_0_0.mp4'
+        return None
     else:
         # Go through all ckpt if possible
         checkpoint_list = sorted(glob(os.path.join(Omegadict['pretrained_model_path'], 'checkpoint_*')))

Tune-A-Video/README.md

@@ -1,119 +0,0 @@
-# Tune-A-Video
-
-This repository is the official implementation of [Tune-A-Video](https://arxiv.org/abs/2212.11565).
-
-**[Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation](https://arxiv.org/abs/2212.11565)**
-<br/>
-[Jay Zhangjie Wu](https://zhangjiewu.github.io/), 
-[Yixiao Ge](https://geyixiao.com/), 
-[Xintao Wang](https://xinntao.github.io/), 
-[Stan Weixian Lei](), 
-[Yuchao Gu](https://ycgu.site/), 
-[Wynne Hsu](https://www.comp.nus.edu.sg/~whsu/), 
-[Ying Shan](https://scholar.google.com/citations?user=4oXBp9UAAAAJ&hl=en), 
-[Xiaohu Qie](https://scholar.google.com/citations?user=mk-F69UAAAAJ&hl=en), 
-[Mike Zheng Shou](https://sites.google.com/view/showlab)
-<br/>
-
-[Project Page](https://tuneavideo.github.io/) | [arXiv](https://arxiv.org/abs/2212.11565)
-
-## Setup
-
-### Requirements
-
-```shell
-pip install -r requirements.txt
-```
-
-Installing [xformers](https://github.com/facebookresearch/xformers) is highly recommended for more efficiency and speed on GPUs. 
-To enable xformers, set `enable_xformers_memory_efficient_attention=True` (default).
-
-### Weights
-
-You can download the pre-trained [Stable Diffusion](https://arxiv.org/abs/2112.10752) models 
-(e.g., [Stable Diffusion v1-4](https://huggingface.co/CompVis/stable-diffusion-v1-4)):
-
-```shell
-git lfs install
-git clone https://huggingface.co/CompVis/stable-diffusion-v1-4
-```
-
-Alternatively, you can use a personalized [DreamBooth](https://arxiv.org/abs/2208.12242) model (e.g., [mr-potato-head](https://huggingface.co/sd-dreambooth-library/mr-potato-head)):
-```shell
-git lfs install
-git clone https://huggingface.co/sd-dreambooth-library/mr-potato-head
-```
-
-## Training
-
-To fine-tune the text-to-image diffusion models for text-to-video generation, run this command:
-
-```shell
-accelerate launch train_tuneavideo.py --config="configs/man-surfing.yaml"
-```
-
-## Inference
-
-Once the training is done, run inference:
-
-```python
-from tuneavideo.pipelines.pipeline_tuneavideo import TuneAVideoPipeline
-from tuneavideo.models.unet import UNet3DConditionModel
-from tuneavideo.util import save_videos_grid
-import torch
-
-model_id = "path-to-your-trained-model"
-unet = UNet3DConditionModel.from_pretrained(model_id, subfolder='unet', torch_dtype=torch.float16).to('cuda')
-pipe = TuneAVideoPipeline.from_pretrained("CompVis/stable-diffusion-v1-4", unet=unet, torch_dtype=torch.float16).to("cuda")
-
-prompt = "a panda is surfing"
-video = pipe(prompt, video_length=8, height=512, width=512, num_inference_steps=50, guidance_scale=7.5).videos
-
-save_videos_grid(video, f"{prompt}.gif")
-```
-
-## Results
-
-### Fine-tuning on Stable Diffusion
-
-<table width="100%" align="center">
-<tr>
-  <td><img src="https://tuneavideo.github.io/static/results/man-surfing/train.gif"></td>
-  <td><img src="https://tuneavideo.github.io/static/results/repo/stablediffusion/panda-surfing.gif"></td>
-  <td><img src="https://tuneavideo.github.io/static/results/repo/stablediffusion/ironman-desert.gif"></td>              
-  <td><img src="https://tuneavideo.github.io/static/results/repo/stablediffusion/raccoon-cartoon.gif"></td>
-</tr>
-<tr>
-  <td width=25% style="text-align:center;color:gray;">[Training] a man is surfing.</td>
-  <td width=25% style="text-align:center;">a panda is surfing.</td>
-  <td width=25% style="text-align:center;">Iron Man is surfing in the desert.</td>
-  <td width=25% style="text-align:center;">a raccoon is surfing, cartoon style.</td>
-</tr>
-</table>
-
-### Fine-tuning on DreamBooth
-
-<table width="100%" align="center">
-<tr>
-  <td><img src="https://tuneavideo.github.io/static/results/repo/dreambooth/mr-potato-head.png"></td>
-  <td><img src="https://tuneavideo.github.io/static/results/repo/dreambooth/pink-hat.gif"></td>
-  <td><img src="https://tuneavideo.github.io/static/results/repo/dreambooth/potato-sunglasses.gif"></td>              
-  <td><img src="https://tuneavideo.github.io/static/results/repo/dreambooth/potato-forest.gif"></td>
-</tr>
-<tr>
-  <td width=25% style="text-align:center;color:gray;">sks mr potato head.</td>
-  <td width=25% style="text-align:center;">sks mr potato head, wearing a pink hat, is surfing.</td>
-  <td width=25% style="text-align:center;">sks mr potato head, wearing sunglasses, is surfing.</td>
-  <td width=25% style="text-align:center;">sks mr potato head is surfing in the forest.</td>
-</tr>
-</table>
-
-## BibTeX
-```
-@article{wu2022tuneavideo,
-    title={Tune-A-Video: One-Shot Tuning of Image Diffusion Models for Text-to-Video Generation},
-    author={Wu, Jay Zhangjie and Ge, Yixiao and Wang, Xintao and Lei, Stan Weixian and Gu, Yuchao and Hsu, Wynne and Shan, Ying and Qie, Xiaohu and Shou, Mike Zheng},
-    journal={arXiv preprint arXiv:2212.11565},
-    year={2022}
-}
-```

Tune-A-Video/configs/man-surfing.yaml

@@ -1,39 +0,0 @@
-pretrained_model_path: "./checkpoints/stable-diffusion-v1-4"
-output_dir: "./outputs/man-surfing_lr3e-5_seed33"
-
-train_data:
-  video_path: "data/man-surfing.mp4"
-  prompt: "a man is surfing"
-  n_sample_frames: 8
-  width: 512
-  height: 512
-  sample_start_idx: 0
-  sample_frame_rate: 1
-
-validation_data:
-  prompts:
-    - "a panda is surfing"
-    - "a boy, wearing a birthday hat, is surfing"
-    - "a raccoon is surfing, cartoon style"
-    - "Iron Man is surfing in the desert"
-  video_length: 8
-  width: 512
-  height: 512
-  num_inference_steps: 50
-  guidance_scale: 7.5
-
-learning_rate: 3e-5
-train_batch_size: 1
-max_train_steps: 300
-checkpointing_steps: 1000
-validation_steps: 100
-trainable_modules:
-  - "attn1.to_q"
-  - "attn2.to_q"
-  - "attn_temp"
-
-seed: 33
-mixed_precision: fp16
-use_8bit_adam: False
-gradient_checkpointing: True
-enable_xformers_memory_efficient_attention: True

Tune-A-Video/configs/mr-potato-head.yaml

@@ -1,39 +0,0 @@
-pretrained_model_path: "./checkpoints/mr-potato-head"
-output_dir: "./outputs/mr-potato-head_lr3e-5_seed33"
-
-train_data:
-  video_path: "data/man-surfing.mp4"
-  prompt: "a man is surfing"
-  n_sample_frames: 8
-  width: 512
-  height: 512
-  sample_start_idx: 0
-  sample_frame_rate: 1
-
-validation_data:
-  prompts:
-    - "sks mr potato head is surfing"
-    - "sks mr potato head, wearing a pink hat, is surfing"
-    - "sks mr potato head, wearing funny sunglasses, is surfing"
-    - "sks mr potato head is surfing in the forest"
-  video_length: 8
-  width: 512
-  height: 512
-  num_inference_steps: 50
-  guidance_scale: 7.5
-
-learning_rate: 3e-5
-train_batch_size: 1
-max_train_steps: 500
-checkpointing_steps: 1000
-validation_steps: 100
-trainable_modules:
-  - "attn1.to_q"
-  - "attn2.to_q"
-  - "attn_temp"
-
-seed: 33
-mixed_precision: fp16
-use_8bit_adam: False
-gradient_checkpointing: True
-enable_xformers_memory_efficient_attention: True

Tune-A-Video/requirements.txt

@@ -1,13 +0,0 @@
-torch==1.12.1
-torchvision==0.13.1
-diffusers[torch]==0.11.1
-transformers>=4.25.1
-bitsandbytes==0.35.4
-decord==0.6.0
-accelerate
-tensorboard
-modelcards
-omegaconf
-einops
-imageio
-ftfy

Tune-A-Video/train_tuneavideo.py

@@ -1,352 +0,0 @@
-import argparse
-import datetime
-import logging
-import inspect
-import math
-import os
-from typing import Dict, Optional, Tuple
-from omegaconf import OmegaConf
-
-import torch
-import torch.nn.functional as F
-import torch.utils.checkpoint
-
-import diffusers
-import transformers
-from accelerate import Accelerator
-from accelerate.logging import get_logger
-from accelerate.utils import set_seed
-from diffusers import AutoencoderKL, DDPMScheduler, DDIMScheduler
-from diffusers.optimization import get_scheduler
-from diffusers.utils import check_min_version
-from diffusers.utils.import_utils import is_xformers_available
-from tqdm.auto import tqdm
-from transformers import CLIPTextModel, CLIPTokenizer
-
-from tuneavideo.models.unet import UNet3DConditionModel
-from tuneavideo.data.dataset import TuneAVideoDataset
-from tuneavideo.pipelines.pipeline_tuneavideo import TuneAVideoPipeline
-from tuneavideo.util import save_videos_grid
-from einops import rearrange
-
-
-# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
-check_min_version("0.10.0.dev0")
-
-logger = get_logger(__name__, log_level="INFO")
-
-
-def main(
-    pretrained_model_path: str,
-    output_dir: str,
-    train_data: Dict,
-    validation_data: Dict,
-    validation_steps: int = 100,
-    trainable_modules: Tuple[str] = (
-        "attn1.to_q",
-        "attn2.to_q",
-        "attn_temp",
-    ),
-    train_batch_size: int = 1,
-    max_train_steps: int = 500,
-    learning_rate: float = 3e-5,
-    scale_lr: bool = False,
-    lr_scheduler: str = "constant",
-    lr_warmup_steps: int = 0,
-    adam_beta1: float = 0.9,
-    adam_beta2: float = 0.999,
-    adam_weight_decay: float = 1e-2,
-    adam_epsilon: float = 1e-08,
-    max_grad_norm: float = 1.0,
-    gradient_accumulation_steps: int = 1,
-    gradient_checkpointing: bool = True,
-    checkpointing_steps: int = 500,
-    resume_from_checkpoint: Optional[str] = None,
-    mixed_precision: Optional[str] = "fp16",
-    use_8bit_adam: bool = False,
-    enable_xformers_memory_efficient_attention: bool = True,
-    seed: Optional[int] = None,
-):
-    *_, config = inspect.getargvalues(inspect.currentframe())
-
-    accelerator = Accelerator(
-        gradient_accumulation_steps=gradient_accumulation_steps,
-        mixed_precision=mixed_precision,
-    )
-
-    # Make one log on every process with the configuration for debugging.
-    logging.basicConfig(
-        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
-        datefmt="%m/%d/%Y %H:%M:%S",
-        level=logging.INFO,
-    )
-    logger.info(accelerator.state, main_process_only=False)
-    if accelerator.is_local_main_process:
-        transformers.utils.logging.set_verbosity_warning()
-        diffusers.utils.logging.set_verbosity_info()
-    else:
-        transformers.utils.logging.set_verbosity_error()
-        diffusers.utils.logging.set_verbosity_error()
-
-    # If passed along, set the training seed now.
-    if seed is not None:
-        set_seed(seed)
-
-    # Handle the output folder creation
-    if accelerator.is_main_process:
-        now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
-        output_dir = os.path.join(output_dir, now)
-        os.makedirs(output_dir, exist_ok=True)
-        OmegaConf.save(config, os.path.join(output_dir, 'config.yaml'))
-
-    # Load scheduler, tokenizer and models.
-    noise_scheduler = DDPMScheduler.from_pretrained(pretrained_model_path, subfolder="scheduler")
-    tokenizer = CLIPTokenizer.from_pretrained(pretrained_model_path, subfolder="tokenizer")
-    text_encoder = CLIPTextModel.from_pretrained(pretrained_model_path, subfolder="text_encoder")
-    vae = AutoencoderKL.from_pretrained(pretrained_model_path, subfolder="vae")
-    unet = UNet3DConditionModel.from_pretrained_2d(pretrained_model_path, subfolder="unet")
-
-    # Freeze vae and text_encoder
-    vae.requires_grad_(False)
-    text_encoder.requires_grad_(False)
-
-    unet.requires_grad_(False)
-    for name, module in unet.named_modules():
-        if name.endswith(tuple(trainable_modules)):
-            for params in module.parameters():
-                params.requires_grad = True
-
-    if enable_xformers_memory_efficient_attention:
-        if is_xformers_available():
-            unet.enable_xformers_memory_efficient_attention()
-        else:
-            raise ValueError("xformers is not available. Make sure it is installed correctly")
-
-    if gradient_checkpointing:
-        unet.enable_gradient_checkpointing()
-
-    if scale_lr:
-        learning_rate = (
-            learning_rate * gradient_accumulation_steps * train_batch_size * accelerator.num_processes
-        )
-
-    # Initialize the optimizer
-    if use_8bit_adam:
-        try:
-            import bitsandbytes as bnb
-        except ImportError:
-            raise ImportError(
-                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
-            )
-
-        optimizer_cls = bnb.optim.AdamW8bit
-    else:
-        optimizer_cls = torch.optim.AdamW
-
-    optimizer = optimizer_cls(
-        unet.parameters(),
-        lr=learning_rate,
-        betas=(adam_beta1, adam_beta2),
-        weight_decay=adam_weight_decay,
-        eps=adam_epsilon,
-    )
-
-    # Get the training dataset
-    train_dataset = TuneAVideoDataset(**train_data)
-
-    # Preprocessing the dataset
-    train_dataset.prompt_ids = tokenizer(
-        train_dataset.prompt, max_length=tokenizer.model_max_length, padding="max_length", truncation=True, return_tensors="pt"
-    ).input_ids[0]
-
-    # DataLoaders creation:
-    train_dataloader = torch.utils.data.DataLoader(
-        train_dataset, batch_size=train_batch_size
-    )
-
-    # Get the validation pipeline
-    validation_pipeline = TuneAVideoPipeline(
-        vae=vae, text_encoder=text_encoder, tokenizer=tokenizer, unet=unet,
-        scheduler=DDIMScheduler.from_pretrained(pretrained_model_path, subfolder="scheduler")
-    )
-
-    # Scheduler
-    lr_scheduler = get_scheduler(
-        lr_scheduler,
-        optimizer=optimizer,
-        num_warmup_steps=lr_warmup_steps * gradient_accumulation_steps,
-        num_training_steps=max_train_steps * gradient_accumulation_steps,
-    )
-
-    # Prepare everything with our `accelerator`.
-    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        unet, optimizer, train_dataloader, lr_scheduler
-    )
-
-    # For mixed precision training we cast the text_encoder and vae weights to half-precision
-    # as these models are only used for inference, keeping weights in full precision is not required.
-    weight_dtype = torch.float32
-    if accelerator.mixed_precision == "fp16":
-        weight_dtype = torch.float16
-    elif accelerator.mixed_precision == "bf16":
-        weight_dtype = torch.bfloat16
-
-    # Move text_encode and vae to gpu and cast to weight_dtype
-    text_encoder.to(accelerator.device, dtype=weight_dtype)
-    vae.to(accelerator.device, dtype=weight_dtype)
-
-    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
-    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / gradient_accumulation_steps)
-    # Afterwards we recalculate our number of training epochs
-    num_train_epochs = math.ceil(max_train_steps / num_update_steps_per_epoch)
-
-    # We need to initialize the trackers we use, and also store our configuration.
-    # The trackers initializes automatically on the main process.
-    if accelerator.is_main_process:
-        accelerator.init_trackers("text2video-fine-tune")
-
-    # Train!
-    total_batch_size = train_batch_size * accelerator.num_processes * gradient_accumulation_steps
-
-    logger.info("***** Running training *****")
-    logger.info(f"  Num examples = {len(train_dataset)}")
-    logger.info(f"  Num Epochs = {num_train_epochs}")
-    logger.info(f"  Instantaneous batch size per device = {train_batch_size}")
-    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
-    logger.info(f"  Gradient Accumulation steps = {gradient_accumulation_steps}")
-    logger.info(f"  Total optimization steps = {max_train_steps}")
-    global_step = 0
-    first_epoch = 0
-
-    # Potentially load in the weights and states from a previous save
-    if resume_from_checkpoint:
-        if resume_from_checkpoint != "latest":
-            path = os.path.basename(resume_from_checkpoint)
-        else:
-            # Get the most recent checkpoint
-            dirs = os.listdir(output_dir)
-            dirs = [d for d in dirs if d.startswith("checkpoint")]
-            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
-            path = dirs[-1]
-        accelerator.print(f"Resuming from checkpoint {path}")
-        accelerator.load_state(os.path.join(output_dir, path))
-        global_step = int(path.split("-")[1])
-
-        first_epoch = global_step // num_update_steps_per_epoch
-        resume_step = global_step % num_update_steps_per_epoch
-
-    # Only show the progress bar once on each machine.
-    progress_bar = tqdm(range(global_step, max_train_steps), disable=not accelerator.is_local_main_process)
-    progress_bar.set_description("Steps")
-
-    for epoch in range(first_epoch, num_train_epochs):
-        unet.train()
-        train_loss = 0.0
-        for step, batch in enumerate(train_dataloader):
-            # Skip steps until we reach the resumed step
-            if resume_from_checkpoint and epoch == first_epoch and step < resume_step:
-                if step % gradient_accumulation_steps == 0:
-                    progress_bar.update(1)
-                continue
-
-            with accelerator.accumulate(unet):
-                # Convert videos to latent space
-                pixel_values = batch["pixel_values"].to(weight_dtype)
-                video_length = pixel_values.shape[1]
-                pixel_values = rearrange(pixel_values, "b f c h w -> (b f) c h w")
-                latents = vae.encode(pixel_values).latent_dist.sample()
-                latents = rearrange(latents, "(b f) c h w -> b c f h w", f=video_length)
-                latents = latents * 0.18215
-
-                # Sample noise that we'll add to the latents
-                noise = torch.randn_like(latents)
-                bsz = latents.shape[0]
-                # Sample a random timestep for each video
-                timesteps = torch.randint(0, noise_scheduler.num_train_timesteps, (bsz,), device=latents.device)
-                timesteps = timesteps.long()
-
-                # Add noise to the latents according to the noise magnitude at each timestep
-                # (this is the forward diffusion process)
-                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
-
-                # Get the text embedding for conditioning
-                encoder_hidden_states = text_encoder(batch["prompt_ids"])[0]
-
-                # Get the target for loss depending on the prediction type
-                if noise_scheduler.prediction_type == "epsilon":
-                    target = noise
-                elif noise_scheduler.prediction_type == "v_prediction":
-                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
-                else:
-                    raise ValueError(f"Unknown prediction type {noise_scheduler.prediction_type}")
-
-                # Predict the noise residual and compute loss
-                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
-                loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
-
-                # Gather the losses across all processes for logging (if we use distributed training).
-                avg_loss = accelerator.gather(loss.repeat(train_batch_size)).mean()
-                train_loss += avg_loss.item() / gradient_accumulation_steps
-
-                # Backpropagate
-                accelerator.backward(loss)
-                if accelerator.sync_gradients:
-                    accelerator.clip_grad_norm_(unet.parameters(), max_grad_norm)
-                optimizer.step()
-                lr_scheduler.step()
-                optimizer.zero_grad()
-
-            # Checks if the accelerator has performed an optimization step behind the scenes
-            if accelerator.sync_gradients:
-                progress_bar.update(1)
-                global_step += 1
-                accelerator.log({"train_loss": train_loss}, step=global_step)
-                train_loss = 0.0
-
-                if global_step % checkpointing_steps == 0:
-                    if accelerator.is_main_process:
-                        save_path = os.path.join(output_dir, f"checkpoint-{global_step}")
-                        accelerator.save_state(save_path)
-                        logger.info(f"Saved state to {save_path}")
-
-                if global_step % validation_steps == 0:
-                    if accelerator.is_main_process:
-                        save_path = os.path.join(output_dir, f"samples/sample-{global_step}.gif")
-                        samples = []
-                        generator = torch.Generator(device=latents.device)
-                        generator.manual_seed(seed)
-                        for idx, prompt in enumerate(validation_data.prompts):
-                            sample = validation_pipeline(prompt, generator=generator, **validation_data).videos
-                            save_videos_grid(sample, os.path.join(output_dir, f"samples/sample-{global_step}/{prompt}.gif"))
-                            samples.append(sample)
-                        samples = torch.concat(samples)
-                        save_videos_grid(samples, save_path)
-                        logger.info(f"Saved samples to {save_path}")
-
-            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
-            progress_bar.set_postfix(**logs)
-
-            if global_step >= max_train_steps:
-                break
-
-    # Create the pipeline using the trained modules and save it.
-    accelerator.wait_for_everyone()
-    if accelerator.is_main_process:
-        unet = accelerator.unwrap_model(unet)
-        pipeline = TuneAVideoPipeline.from_pretrained(
-            pretrained_model_path,
-            text_encoder=text_encoder,
-            vae=vae,
-            unet=unet,
-        )
-        pipeline.save_pretrained(output_dir)
-
-    accelerator.end_training()
-
-
-if __name__ == "__main__":
-    parser = argparse.ArgumentParser()
-    parser.add_argument("--config", type=str, default="./configs/tuneavideo.yaml")
-    args = parser.parse_args()
-
-    main(**OmegaConf.load(args.config))

Tune-A-Video/tuneavideo/data/dataset.py

@@ -1,44 +0,0 @@
-import decord
-decord.bridge.set_bridge('torch')
-
-from torch.utils.data import Dataset
-from einops import rearrange
-
-
-class TuneAVideoDataset(Dataset):
-    def __init__(
-            self,
-            video_path: str,
-            prompt: str,
-            width: int = 512,
-            height: int = 512,
-            n_sample_frames: int = 8,
-            sample_start_idx: int = 0,
-            sample_frame_rate: int = 1,
-    ):
-        self.video_path = video_path
-        self.prompt = prompt
-        self.prompt_ids = None
-
-        self.width = width
-        self.height = height
-        self.n_sample_frames = n_sample_frames
-        self.sample_start_idx = sample_start_idx
-        self.sample_frame_rate = sample_frame_rate
-
-    def __len__(self):
-        return 1
-
-    def __getitem__(self, index):
-        # load and sample video frames
-        vr = decord.VideoReader(self.video_path, width=self.width, height=self.height)
-        sample_index = list(range(self.sample_start_idx, len(vr), self.sample_frame_rate))[:self.n_sample_frames]
-        video = vr.get_batch(sample_index)
-        video = rearrange(video, "f h w c -> f c h w")
-
-        example = {
-            "pixel_values": (video / 127.5 - 1.0),
-            "prompt_ids": self.prompt_ids
-        }
-
-        return example

Tune-A-Video/tuneavideo/models/attention.py

@@ -1,328 +0,0 @@
-# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py
-
-from dataclasses import dataclass
-from typing import Optional
-
-import torch
-import torch.nn.functional as F
-from torch import nn
-
-from diffusers.configuration_utils import ConfigMixin, register_to_config
-from diffusers.modeling_utils import ModelMixin
-from diffusers.utils import BaseOutput
-from diffusers.utils.import_utils import is_xformers_available
-from diffusers.models.attention import CrossAttention, FeedForward, AdaLayerNorm
-
-from einops import rearrange, repeat
-
-
-@dataclass
-class Transformer3DModelOutput(BaseOutput):
-    sample: torch.FloatTensor
-
-
-if is_xformers_available():
-    import xformers
-    import xformers.ops
-else:
-    xformers = None
-
-
-class Transformer3DModel(ModelMixin, ConfigMixin):
-    @register_to_config
-    def __init__(
-        self,
-        num_attention_heads: int = 16,
-        attention_head_dim: int = 88,
-        in_channels: Optional[int] = None,
-        num_layers: int = 1,
-        dropout: float = 0.0,
-        norm_num_groups: int = 32,
-        cross_attention_dim: Optional[int] = None,
-        attention_bias: bool = False,
-        activation_fn: str = "geglu",
-        num_embeds_ada_norm: Optional[int] = None,
-        use_linear_projection: bool = False,
-        only_cross_attention: bool = False,
-        upcast_attention: bool = False,
-    ):
-        super().__init__()
-        self.use_linear_projection = use_linear_projection
-        self.num_attention_heads = num_attention_heads
-        self.attention_head_dim = attention_head_dim
-        inner_dim = num_attention_heads * attention_head_dim
-
-        # Define input layers
-        self.in_channels = in_channels
-
-        self.norm = torch.nn.GroupNorm(num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True)
-        if use_linear_projection:
-            self.proj_in = nn.Linear(in_channels, inner_dim)
-        else:
-            self.proj_in = nn.Conv2d(in_channels, inner_dim, kernel_size=1, stride=1, padding=0)
-
-        # Define transformers blocks
-        self.transformer_blocks = nn.ModuleList(
-            [
-                BasicTransformerBlock(
-                    inner_dim,
-                    num_attention_heads,
-                    attention_head_dim,
-                    dropout=dropout,
-                    cross_attention_dim=cross_attention_dim,
-                    activation_fn=activation_fn,
-                    num_embeds_ada_norm=num_embeds_ada_norm,
-                    attention_bias=attention_bias,
-                    only_cross_attention=only_cross_attention,
-                    upcast_attention=upcast_attention,
-                )
-                for d in range(num_layers)
-            ]
-        )
-
-        # 4. Define output layers
-        if use_linear_projection:
-            self.proj_out = nn.Linear(in_channels, inner_dim)
-        else:
-            self.proj_out = nn.Conv2d(inner_dim, in_channels, kernel_size=1, stride=1, padding=0)
-
-    def forward(self, hidden_states, encoder_hidden_states=None, timestep=None, return_dict: bool = True):
-        # Input
-        assert hidden_states.dim() == 5, f"Expected hidden_states to have ndim=5, but got ndim={hidden_states.dim()}."
-        video_length = hidden_states.shape[2]
-        hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w")
-        encoder_hidden_states = repeat(encoder_hidden_states, 'b n c -> (b f) n c', f=video_length)
-
-        batch, channel, height, weight = hidden_states.shape
-        residual = hidden_states
-
-        hidden_states = self.norm(hidden_states)
-        if not self.use_linear_projection:
-            hidden_states = self.proj_in(hidden_states)
-            inner_dim = hidden_states.shape[1]
-            hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * weight, inner_dim)
-        else:
-            inner_dim = hidden_states.shape[1]
-            hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(batch, height * weight, inner_dim)
-            hidden_states = self.proj_in(hidden_states)
-
-        # Blocks
-        for block in self.transformer_blocks:
-            hidden_states = block(
-                hidden_states,
-                encoder_hidden_states=encoder_hidden_states,
-                timestep=timestep,
-                video_length=video_length
-            )
-
-        # Output
-        if not self.use_linear_projection:
-            hidden_states = (
-                hidden_states.reshape(batch, height, weight, inner_dim).permute(0, 3, 1, 2).contiguous()
-            )
-            hidden_states = self.proj_out(hidden_states)
-        else:
-            hidden_states = self.proj_out(hidden_states)
-            hidden_states = (
-                hidden_states.reshape(batch, height, weight, inner_dim).permute(0, 3, 1, 2).contiguous()
-            )
-
-        output = hidden_states + residual
-
-        output = rearrange(output, "(b f) c h w -> b c f h w", f=video_length)
-        if not return_dict:
-            return (output,)
-
-        return Transformer3DModelOutput(sample=output)
-
-
-class BasicTransformerBlock(nn.Module):
-    def __init__(
-        self,
-        dim: int,
-        num_attention_heads: int,
-        attention_head_dim: int,
-        dropout=0.0,
-        cross_attention_dim: Optional[int] = None,
-        activation_fn: str = "geglu",
-        num_embeds_ada_norm: Optional[int] = None,
-        attention_bias: bool = False,
-        only_cross_attention: bool = False,
-        upcast_attention: bool = False,
-    ):
-        super().__init__()
-        self.only_cross_attention = only_cross_attention
-        self.use_ada_layer_norm = num_embeds_ada_norm is not None
-
-        # SC-Attn
-        self.attn1 = SparseCausalAttention(
-            query_dim=dim,
-            heads=num_attention_heads,
-            dim_head=attention_head_dim,
-            dropout=dropout,
-            bias=attention_bias,
-            cross_attention_dim=cross_attention_dim if only_cross_attention else None,
-            upcast_attention=upcast_attention,
-        )
-        self.norm1 = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
-
-        # Cross-Attn
-        if cross_attention_dim is not None:
-            self.attn2 = CrossAttention(
-                query_dim=dim,
-                cross_attention_dim=cross_attention_dim,
-                heads=num_attention_heads,
-                dim_head=attention_head_dim,
-                dropout=dropout,
-                bias=attention_bias,
-                upcast_attention=upcast_attention,
-            )
-        else:
-            self.attn2 = None
-
-        if cross_attention_dim is not None:
-            self.norm2 = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
-        else:
-            self.norm2 = None
-
-        # Feed-forward
-        self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn)
-        self.norm3 = nn.LayerNorm(dim)
-
-        # Temp-Attn
-        self.attn_temp = CrossAttention(
-            query_dim=dim,
-            heads=num_attention_heads,
-            dim_head=attention_head_dim,
-            dropout=dropout,
-            bias=attention_bias,
-            upcast_attention=upcast_attention,
-        )
-        nn.init.zeros_(self.attn_temp.to_out[0].weight.data)
-        self.norm_temp = AdaLayerNorm(dim, num_embeds_ada_norm) if self.use_ada_layer_norm else nn.LayerNorm(dim)
-
-    def set_use_memory_efficient_attention_xformers(self, use_memory_efficient_attention_xformers: bool):
-        if not is_xformers_available():
-            print("Here is how to install it")
-            raise ModuleNotFoundError(
-                "Refer to https://github.com/facebookresearch/xformers for more information on how to install"
-                " xformers",
-                name="xformers",
-            )
-        elif not torch.cuda.is_available():
-            raise ValueError(
-                "torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is only"
-                " available for GPU "
-            )
-        else:
-            try:
-                # Make sure we can run the memory efficient attention
-                _ = xformers.ops.memory_efficient_attention(
-                    torch.randn((1, 2, 40), device="cuda"),
-                    torch.randn((1, 2, 40), device="cuda"),
-                    torch.randn((1, 2, 40), device="cuda"),
-                )
-            except Exception as e:
-                raise e
-            self.attn1._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
-            if self.attn2 is not None:
-                self.attn2._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
-            # self.attn_temp._use_memory_efficient_attention_xformers = use_memory_efficient_attention_xformers
-
-    def forward(self, hidden_states, encoder_hidden_states=None, timestep=None, attention_mask=None, video_length=None):
-        # SparseCausal-Attention
-        norm_hidden_states = (
-            self.norm1(hidden_states, timestep) if self.use_ada_layer_norm else self.norm1(hidden_states)
-        )
-
-        if self.only_cross_attention:
-            hidden_states = (
-                self.attn1(norm_hidden_states, encoder_hidden_states, attention_mask=attention_mask) + hidden_states
-            )
-        else:
-            hidden_states = self.attn1(norm_hidden_states, attention_mask=attention_mask, video_length=video_length) + hidden_states
-
-        if self.attn2 is not None:
-            # Cross-Attention
-            norm_hidden_states = (
-                self.norm2(hidden_states, timestep) if self.use_ada_layer_norm else self.norm2(hidden_states)
-            )
-            hidden_states = (
-                self.attn2(
-                    norm_hidden_states, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask
-                )
-                + hidden_states
-            )
-
-        # Feed-forward
-        hidden_states = self.ff(self.norm3(hidden_states)) + hidden_states
-
-        # Temporal-Attention
-        d = hidden_states.shape[1]
-        hidden_states = rearrange(hidden_states, "(b f) d c -> (b d) f c", f=video_length)
-        norm_hidden_states = (
-            self.norm_temp(hidden_states, timestep) if self.use_ada_layer_norm else self.norm_temp(hidden_states)
-        )
-        hidden_states = self.attn_temp(norm_hidden_states) + hidden_states
-        hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d)
-
-        return hidden_states
-
-
-class SparseCausalAttention(CrossAttention):
-    def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None, video_length=None):
-        batch_size, sequence_length, _ = hidden_states.shape
-
-        encoder_hidden_states = encoder_hidden_states
-
-        if self.group_norm is not None:
-            hidden_states = self.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
-
-        query = self.to_q(hidden_states)
-        dim = query.shape[-1]
-        query = self.reshape_heads_to_batch_dim(query)
-
-        if self.added_kv_proj_dim is not None:
-            raise NotImplementedError
-
-        encoder_hidden_states = encoder_hidden_states if encoder_hidden_states is not None else hidden_states
-        key = self.to_k(encoder_hidden_states)
-        value = self.to_v(encoder_hidden_states)
-
-        former_frame_index = torch.arange(video_length) - 1
-        former_frame_index[0] = 0
-
-        key = rearrange(key, "(b f) d c -> b f d c", f=video_length)
-        key = torch.cat([key[:, [0] * video_length], key[:, former_frame_index]], dim=2)
-        key = rearrange(key, "b f d c -> (b f) d c")
-
-        value = rearrange(value, "(b f) d c -> b f d c", f=video_length)
-        value = torch.cat([value[:, [0] * video_length], value[:, former_frame_index]], dim=2)
-        value = rearrange(value, "b f d c -> (b f) d c")
-
-        key = self.reshape_heads_to_batch_dim(key)
-        value = self.reshape_heads_to_batch_dim(value)
-
-        if attention_mask is not None:
-            if attention_mask.shape[-1] != query.shape[1]:
-                target_length = query.shape[1]
-                attention_mask = F.pad(attention_mask, (0, target_length), value=0.0)
-                attention_mask = attention_mask.repeat_interleave(self.heads, dim=0)
-
-        # attention, what we cannot get enough of
-        if self._use_memory_efficient_attention_xformers:
-            hidden_states = self._memory_efficient_attention_xformers(query, key, value, attention_mask)
-            # Some versions of xformers return output in fp32, cast it back to the dtype of the input
-            hidden_states = hidden_states.to(query.dtype)
-        else:
-            if self._slice_size is None or query.shape[0] // self._slice_size == 1:
-                hidden_states = self._attention(query, key, value, attention_mask)
-            else:
-                hidden_states = self._sliced_attention(query, key, value, sequence_length, dim, attention_mask)
-
-        # linear proj
-        hidden_states = self.to_out[0](hidden_states)
-
-        # dropout
-        hidden_states = self.to_out[1](hidden_states)
-        return hidden_states

Tune-A-Video/tuneavideo/models/resnet.py

@@ -1,209 +0,0 @@
-# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py
-
-import torch
-import torch.nn as nn
-import torch.nn.functional as F
-
-from einops import rearrange
-
-
-class InflatedConv3d(nn.Conv2d):
-    def forward(self, x):
-        video_length = x.shape[2]
-
-        x = rearrange(x, "b c f h w -> (b f) c h w")
-        x = super().forward(x)
-        x = rearrange(x, "(b f) c h w -> b c f h w", f=video_length)
-
-        return x
-
-
-class Upsample3D(nn.Module):
-    def __init__(self, channels, use_conv=False, use_conv_transpose=False, out_channels=None, name="conv"):
-        super().__init__()
-        self.channels = channels
-        self.out_channels = out_channels or channels
-        self.use_conv = use_conv
-        self.use_conv_transpose = use_conv_transpose
-        self.name = name
-
-        conv = None
-        if use_conv_transpose:
-            raise NotImplementedError
-        elif use_conv:
-            conv = InflatedConv3d(self.channels, self.out_channels, 3, padding=1)
-
-        if name == "conv":
-            self.conv = conv
-        else:
-            self.Conv2d_0 = conv
-
-    def forward(self, hidden_states, output_size=None):
-        assert hidden_states.shape[1] == self.channels
-
-        if self.use_conv_transpose:
-            raise NotImplementedError
-
-        # Cast to float32 to as 'upsample_nearest2d_out_frame' op does not support bfloat16
-        dtype = hidden_states.dtype
-        if dtype == torch.bfloat16:
-            hidden_states = hidden_states.to(torch.float32)
-
-        # upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
-        if hidden_states.shape[0] >= 64:
-            hidden_states = hidden_states.contiguous()
-
-        # if `output_size` is passed we force the interpolation output
-        # size and do not make use of `scale_factor=2`
-        if output_size is None:
-            hidden_states = F.interpolate(hidden_states, scale_factor=[1.0, 2.0, 2.0], mode="nearest")
-        else:
-            hidden_states = F.interpolate(hidden_states, size=output_size, mode="nearest")
-
-        # If the input is bfloat16, we cast back to bfloat16
-        if dtype == torch.bfloat16:
-            hidden_states = hidden_states.to(dtype)
-
-        if self.use_conv:
-            if self.name == "conv":
-                hidden_states = self.conv(hidden_states)
-            else:
-                hidden_states = self.Conv2d_0(hidden_states)
-
-        return hidden_states
-
-
-class Downsample3D(nn.Module):
-    def __init__(self, channels, use_conv=False, out_channels=None, padding=1, name="conv"):
-        super().__init__()
-        self.channels = channels
-        self.out_channels = out_channels or channels
-        self.use_conv = use_conv
-        self.padding = padding
-        stride = 2
-        self.name = name
-
-        if use_conv:
-            conv = InflatedConv3d(self.channels, self.out_channels, 3, stride=stride, padding=padding)
-        else:
-            raise NotImplementedError
-
-        if name == "conv":
-            self.Conv2d_0 = conv
-            self.conv = conv
-        elif name == "Conv2d_0":
-            self.conv = conv
-        else:
-            self.conv = conv
-
-    def forward(self, hidden_states):
-        assert hidden_states.shape[1] == self.channels
-        if self.use_conv and self.padding == 0:
-            raise NotImplementedError
-
-        assert hidden_states.shape[1] == self.channels
-        hidden_states = self.conv(hidden_states)
-
-        return hidden_states
-
-
-class ResnetBlock3D(nn.Module):
-    def __init__(
-        self,
-        *,
-        in_channels,
-        out_channels=None,
-        conv_shortcut=False,
-        dropout=0.0,
-        temb_channels=512,
-        groups=32,
-        groups_out=None,
-        pre_norm=True,
-        eps=1e-6,
-        non_linearity="swish",
-        time_embedding_norm="default",
-        output_scale_factor=1.0,
-        use_in_shortcut=None,
-    ):
-        super().__init__()
-        self.pre_norm = pre_norm
-        self.pre_norm = True
-        self.in_channels = in_channels
-        out_channels = in_channels if out_channels is None else out_channels
-        self.out_channels = out_channels
-        self.use_conv_shortcut = conv_shortcut
-        self.time_embedding_norm = time_embedding_norm
-        self.output_scale_factor = output_scale_factor
-
-        if groups_out is None:
-            groups_out = groups
-
-        self.norm1 = torch.nn.GroupNorm(num_groups=groups, num_channels=in_channels, eps=eps, affine=True)
-
-        self.conv1 = InflatedConv3d(in_channels, out_channels, kernel_size=3, stride=1, padding=1)
-
-        if temb_channels is not None:
-            if self.time_embedding_norm == "default":
-                time_emb_proj_out_channels = out_channels
-            elif self.time_embedding_norm == "scale_shift":
-                time_emb_proj_out_channels = out_channels * 2
-            else:
-                raise ValueError(f"unknown time_embedding_norm : {self.time_embedding_norm} ")
-
-            self.time_emb_proj = torch.nn.Linear(temb_channels, time_emb_proj_out_channels)
-        else:
-            self.time_emb_proj = None
-
-        self.norm2 = torch.nn.GroupNorm(num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True)
-        self.dropout = torch.nn.Dropout(dropout)
-        self.conv2 = InflatedConv3d(out_channels, out_channels, kernel_size=3, stride=1, padding=1)
-
-        if non_linearity == "swish":
-            self.nonlinearity = lambda x: F.silu(x)
-        elif non_linearity == "mish":
-            self.nonlinearity = Mish()
-        elif non_linearity == "silu":
-            self.nonlinearity = nn.SiLU()
-
-        self.use_in_shortcut = self.in_channels != self.out_channels if use_in_shortcut is None else use_in_shortcut
-
-        self.conv_shortcut = None
-        if self.use_in_shortcut:
-            self.conv_shortcut = InflatedConv3d(in_channels, out_channels, kernel_size=1, stride=1, padding=0)
-
-    def forward(self, input_tensor, temb):
-        hidden_states = input_tensor
-
-        hidden_states = self.norm1(hidden_states)
-        hidden_states = self.nonlinearity(hidden_states)
-
-        hidden_states = self.conv1(hidden_states)
-
-        if temb is not None:
-            temb = self.time_emb_proj(self.nonlinearity(temb))[:, :, None, None, None]
-
-        if temb is not None and self.time_embedding_norm == "default":
-            hidden_states = hidden_states + temb
-
-        hidden_states = self.norm2(hidden_states)
-
-        if temb is not None and self.time_embedding_norm == "scale_shift":
-            scale, shift = torch.chunk(temb, 2, dim=1)
-            hidden_states = hidden_states * (1 + scale) + shift
-
-        hidden_states = self.nonlinearity(hidden_states)
-
-        hidden_states = self.dropout(hidden_states)
-        hidden_states = self.conv2(hidden_states)
-
-        if self.conv_shortcut is not None:
-            input_tensor = self.conv_shortcut(input_tensor)
-
-        output_tensor = (input_tensor + hidden_states) / self.output_scale_factor
-
-        return output_tensor
-
-
-class Mish(torch.nn.Module):
-    def forward(self, hidden_states):
-        return hidden_states * torch.tanh(torch.nn.functional.softplus(hidden_states))

Tune-A-Video/tuneavideo/models/unet.py

@@ -1,450 +0,0 @@
-# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_condition.py
-
-from dataclasses import dataclass
-from typing import List, Optional, Tuple, Union
-
-import os
-import json
-
-import torch
-import torch.nn as nn
-import torch.utils.checkpoint
-
-from diffusers.configuration_utils import ConfigMixin, register_to_config
-from diffusers.modeling_utils import ModelMixin
-from diffusers.utils import BaseOutput, logging
-from diffusers.models.embeddings import TimestepEmbedding, Timesteps
-from .unet_blocks import (
-    CrossAttnDownBlock3D,
-    CrossAttnUpBlock3D,
-    DownBlock3D,
-    UNetMidBlock3DCrossAttn,
-    UpBlock3D,
-    get_down_block,
-    get_up_block,
-)
-from .resnet import InflatedConv3d
-
-
-logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
-
-
-@dataclass
-class UNet3DConditionOutput(BaseOutput):
-    sample: torch.FloatTensor
-
-
-class UNet3DConditionModel(ModelMixin, ConfigMixin):
-    _supports_gradient_checkpointing = True
-
-    @register_to_config
-    def __init__(
-        self,
-        sample_size: Optional[int] = None,
-        in_channels: int = 4,
-        out_channels: int = 4,
-        center_input_sample: bool = False,
-        flip_sin_to_cos: bool = True,
-        freq_shift: int = 0,
-        down_block_types: Tuple[str] = (
-            "CrossAttnDownBlock3D",
-            "CrossAttnDownBlock3D",
-            "CrossAttnDownBlock3D",
-            "DownBlock3D",
-        ),
-        mid_block_type: str = "UNetMidBlock3DCrossAttn",
-        up_block_types: Tuple[str] = (
-            "UpBlock3D",
-            "CrossAttnUpBlock3D",
-            "CrossAttnUpBlock3D",
-            "CrossAttnUpBlock3D"
-        ),
-        only_cross_attention: Union[bool, Tuple[bool]] = False,
-        block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
-        layers_per_block: int = 2,
-        downsample_padding: int = 1,
-        mid_block_scale_factor: float = 1,
-        act_fn: str = "silu",
-        norm_num_groups: int = 32,
-        norm_eps: float = 1e-5,
-        cross_attention_dim: int = 1280,
-        attention_head_dim: Union[int, Tuple[int]] = 8,
-        dual_cross_attention: bool = False,
-        use_linear_projection: bool = False,
-        class_embed_type: Optional[str] = None,
-        num_class_embeds: Optional[int] = None,
-        upcast_attention: bool = False,
-        resnet_time_scale_shift: str = "default",
-    ):
-        super().__init__()
-
-        self.sample_size = sample_size
-        time_embed_dim = block_out_channels[0] * 4
-
-        # input
-        self.conv_in = InflatedConv3d(in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1))
-
-        # time
-        self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
-        timestep_input_dim = block_out_channels[0]
-
-        self.time_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
-
-        # class embedding
-        if class_embed_type is None and num_class_embeds is not None:
-            self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
-        elif class_embed_type == "timestep":
-            self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
-        elif class_embed_type == "identity":
-            self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
-        else:
-            self.class_embedding = None
-
-        self.down_blocks = nn.ModuleList([])
-        self.mid_block = None
-        self.up_blocks = nn.ModuleList([])
-
-        if isinstance(only_cross_attention, bool):
-            only_cross_attention = [only_cross_attention] * len(down_block_types)
-
-        if isinstance(attention_head_dim, int):
-            attention_head_dim = (attention_head_dim,) * len(down_block_types)
-
-        # down
-        output_channel = block_out_channels[0]
-        for i, down_block_type in enumerate(down_block_types):
-            input_channel = output_channel
-            output_channel = block_out_channels[i]
-            is_final_block = i == len(block_out_channels) - 1
-
-            down_block = get_down_block(
-                down_block_type,
-                num_layers=layers_per_block,
-                in_channels=input_channel,
-                out_channels=output_channel,
-                temb_channels=time_embed_dim,
-                add_downsample=not is_final_block,
-                resnet_eps=norm_eps,
-                resnet_act_fn=act_fn,
-                resnet_groups=norm_num_groups,
-                cross_attention_dim=cross_attention_dim,
-                attn_num_head_channels=attention_head_dim[i],
-                downsample_padding=downsample_padding,
-                dual_cross_attention=dual_cross_attention,
-                use_linear_projection=use_linear_projection,
-                only_cross_attention=only_cross_attention[i],
-                upcast_attention=upcast_attention,
-                resnet_time_scale_shift=resnet_time_scale_shift,
-            )
-            self.down_blocks.append(down_block)
-
-        # mid
-        if mid_block_type == "UNetMidBlock3DCrossAttn":
-            self.mid_block = UNetMidBlock3DCrossAttn(
-                in_channels=block_out_channels[-1],
-                temb_channels=time_embed_dim,
-                resnet_eps=norm_eps,
-                resnet_act_fn=act_fn,
-                output_scale_factor=mid_block_scale_factor,
-                resnet_time_scale_shift=resnet_time_scale_shift,
-                cross_attention_dim=cross_attention_dim,
-                attn_num_head_channels=attention_head_dim[-1],
-                resnet_groups=norm_num_groups,
-                dual_cross_attention=dual_cross_attention,
-                use_linear_projection=use_linear_projection,
-                upcast_attention=upcast_attention,
-            )
-        else:
-            raise ValueError(f"unknown mid_block_type : {mid_block_type}")
-
-        # count how many layers upsample the videos
-        self.num_upsamplers = 0
-
-        # up
-        reversed_block_out_channels = list(reversed(block_out_channels))
-        reversed_attention_head_dim = list(reversed(attention_head_dim))
-        only_cross_attention = list(reversed(only_cross_attention))
-        output_channel = reversed_block_out_channels[0]
-        for i, up_block_type in enumerate(up_block_types):
-            is_final_block = i == len(block_out_channels) - 1
-
-            prev_output_channel = output_channel
-            output_channel = reversed_block_out_channels[i]
-            input_channel = reversed_block_out_channels[min(i + 1, len(block_out_channels) - 1)]
-
-            # add upsample block for all BUT final layer
-            if not is_final_block:
-                add_upsample = True
-                self.num_upsamplers += 1
-            else:
-                add_upsample = False
-
-            up_block = get_up_block(
-                up_block_type,
-                num_layers=layers_per_block + 1,
-                in_channels=input_channel,
-                out_channels=output_channel,
-                prev_output_channel=prev_output_channel,
-                temb_channels=time_embed_dim,
-                add_upsample=add_upsample,
-                resnet_eps=norm_eps,
-                resnet_act_fn=act_fn,
-                resnet_groups=norm_num_groups,
-                cross_attention_dim=cross_attention_dim,
-                attn_num_head_channels=reversed_attention_head_dim[i],
-                dual_cross_attention=dual_cross_attention,
-                use_linear_projection=use_linear_projection,
-                only_cross_attention=only_cross_attention[i],
-                upcast_attention=upcast_attention,
-                resnet_time_scale_shift=resnet_time_scale_shift,
-            )
-            self.up_blocks.append(up_block)
-            prev_output_channel = output_channel
-
-        # out
-        self.conv_norm_out = nn.GroupNorm(num_channels=block_out_channels[0], num_groups=norm_num_groups, eps=norm_eps)
-        self.conv_act = nn.SiLU()
-        self.conv_out = InflatedConv3d(block_out_channels[0], out_channels, kernel_size=3, padding=1)
-
-    def set_attention_slice(self, slice_size):
-        r"""
-        Enable sliced attention computation.
-
-        When this option is enabled, the attention module will split the input tensor in slices, to compute attention
-        in several steps. This is useful to save some memory in exchange for a small speed decrease.
-
-        Args:
-            slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
-                When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
-                `"max"`, maxium amount of memory will be saved by running only one slice at a time. If a number is
-                provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
-                must be a multiple of `slice_size`.
-        """
-        sliceable_head_dims = []
-
-        def fn_recursive_retrieve_slicable_dims(module: torch.nn.Module):
-            if hasattr(module, "set_attention_slice"):
-                sliceable_head_dims.append(module.sliceable_head_dim)
-
-            for child in module.children():
-                fn_recursive_retrieve_slicable_dims(child)
-
-        # retrieve number of attention layers
-        for module in self.children():
-            fn_recursive_retrieve_slicable_dims(module)
-
-        num_slicable_layers = len(sliceable_head_dims)
-
-        if slice_size == "auto":
-            # half the attention head size is usually a good trade-off between
-            # speed and memory
-            slice_size = [dim // 2 for dim in sliceable_head_dims]
-        elif slice_size == "max":
-            # make smallest slice possible
-            slice_size = num_slicable_layers * [1]
-
-        slice_size = num_slicable_layers * [slice_size] if not isinstance(slice_size, list) else slice_size
-
-        if len(slice_size) != len(sliceable_head_dims):
-            raise ValueError(
-                f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
-                f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
-            )
-
-        for i in range(len(slice_size)):
-            size = slice_size[i]
-            dim = sliceable_head_dims[i]
-            if size is not None and size > dim:
-                raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
-
-        # Recursively walk through all the children.
-        # Any children which exposes the set_attention_slice method
-        # gets the message
-        def fn_recursive_set_attention_slice(module: torch.nn.Module, slice_size: List[int]):
-            if hasattr(module, "set_attention_slice"):
-                module.set_attention_slice(slice_size.pop())
-
-            for child in module.children():
-                fn_recursive_set_attention_slice(child, slice_size)
-
-        reversed_slice_size = list(reversed(slice_size))
-        for module in self.children():
-            fn_recursive_set_attention_slice(module, reversed_slice_size)
-
-    def _set_gradient_checkpointing(self, module, value=False):
-        if isinstance(module, (CrossAttnDownBlock3D, DownBlock3D, CrossAttnUpBlock3D, UpBlock3D)):
-            module.gradient_checkpointing = value
-
-    def forward(
-        self,
-        sample: torch.FloatTensor,
-        timestep: Union[torch.Tensor, float, int],
-        encoder_hidden_states: torch.Tensor,
-        class_labels: Optional[torch.Tensor] = None,
-        attention_mask: Optional[torch.Tensor] = None,
-        return_dict: bool = True,
-    ) -> Union[UNet3DConditionOutput, Tuple]:
-        r"""
-        Args:
-            sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
-            timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps
-            encoder_hidden_states (`torch.FloatTensor`): (batch, sequence_length, feature_dim) encoder hidden states
-            return_dict (`bool`, *optional*, defaults to `True`):
-                Whether or not to return a [`models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain tuple.
-
-        Returns:
-            [`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
-            [`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When
-            returning a tuple, the first element is the sample tensor.
-        """
-        # By default samples have to be AT least a multiple of the overall upsampling factor.
-        # The overall upsampling factor is equal to 2 ** (# num of upsampling layears).
-        # However, the upsampling interpolation output size can be forced to fit any upsampling size
-        # on the fly if necessary.
-        default_overall_up_factor = 2**self.num_upsamplers
-
-        # upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
-        forward_upsample_size = False
-        upsample_size = None
-
-        if any(s % default_overall_up_factor != 0 for s in sample.shape[-2:]):
-            logger.info("Forward upsample size to force interpolation output size.")
-            forward_upsample_size = True
-
-        # prepare attention_mask
-        if attention_mask is not None:
-            attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
-            attention_mask = attention_mask.unsqueeze(1)
-
-        # center input if necessary
-        if self.config.center_input_sample:
-            sample = 2 * sample - 1.0
-
-        # time
-        timesteps = timestep
-        if not torch.is_tensor(timesteps):
-            # This would be a good case for the `match` statement (Python 3.10+)
-            is_mps = sample.device.type == "mps"
-            if isinstance(timestep, float):
-                dtype = torch.float32 if is_mps else torch.float64
-            else:
-                dtype = torch.int32 if is_mps else torch.int64
-            timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
-        elif len(timesteps.shape) == 0:
-            timesteps = timesteps[None].to(sample.device)
-
-        # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
-        timesteps = timesteps.expand(sample.shape[0])
-
-        t_emb = self.time_proj(timesteps)
-
-        # timesteps does not contain any weights and will always return f32 tensors
-        # but time_embedding might actually be running in fp16. so we need to cast here.
-        # there might be better ways to encapsulate this.
-        t_emb = t_emb.to(dtype=self.dtype)
-        emb = self.time_embedding(t_emb)
-
-        if self.class_embedding is not None:
-            if class_labels is None:
-                raise ValueError("class_labels should be provided when num_class_embeds > 0")
-
-            if self.config.class_embed_type == "timestep":
-                class_labels = self.time_proj(class_labels)
-
-            class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
-            emb = emb + class_emb
-
-        # pre-process
-        sample = self.conv_in(sample)
-
-        # down
-        down_block_res_samples = (sample,)
-        for downsample_block in self.down_blocks:
-            if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
-                sample, res_samples = downsample_block(
-                    hidden_states=sample,
-                    temb=emb,
-                    encoder_hidden_states=encoder_hidden_states,
-                    attention_mask=attention_mask,
-                )
-            else:
-                sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
-
-            down_block_res_samples += res_samples
-
-        # mid
-        sample = self.mid_block(
-            sample, emb, encoder_hidden_states=encoder_hidden_states, attention_mask=attention_mask
-        )
-
-        # up
-        for i, upsample_block in enumerate(self.up_blocks):
-            is_final_block = i == len(self.up_blocks) - 1
-
-            res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
-            down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
-
-            # if we have not reached the final block and need to forward the
-            # upsample size, we do it here
-            if not is_final_block and forward_upsample_size:
-                upsample_size = down_block_res_samples[-1].shape[2:]
-
-            if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
-                sample = upsample_block(
-                    hidden_states=sample,
-                    temb=emb,
-                    res_hidden_states_tuple=res_samples,
-                    encoder_hidden_states=encoder_hidden_states,
-                    upsample_size=upsample_size,
-                    attention_mask=attention_mask,
-                )
-            else:
-                sample = upsample_block(
-                    hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples, upsample_size=upsample_size
-                )
-        # post-process
-        sample = self.conv_norm_out(sample)
-        sample = self.conv_act(sample)
-        sample = self.conv_out(sample)
-
-        if not return_dict:
-            return (sample,)
-
-        return UNet3DConditionOutput(sample=sample)
-
-    @classmethod
-    def from_pretrained_2d(cls, pretrained_model_path, subfolder=None):
-        if subfolder is not None:
-            pretrained_model_path = os.path.join(pretrained_model_path, subfolder)
-
-        config_file = os.path.join(pretrained_model_path, 'config.json')
-        if not os.path.isfile(config_file):
-            raise RuntimeError(f"{config_file} does not exist")
-        with open(config_file, "r") as f:
-            config = json.load(f)
-        config["_class_name"] = cls.__name__
-        config["down_block_types"] = [
-            "CrossAttnDownBlock3D",
-            "CrossAttnDownBlock3D",
-            "CrossAttnDownBlock3D",
-            "DownBlock3D"
-        ]
-        config["up_block_types"] = [
-            "UpBlock3D",
-            "CrossAttnUpBlock3D",
-            "CrossAttnUpBlock3D",
-            "CrossAttnUpBlock3D"
-        ]
-
-        from diffusers.utils import WEIGHTS_NAME
-        model = cls.from_config(config)
-        model_file = os.path.join(pretrained_model_path, WEIGHTS_NAME)
-        if not os.path.isfile(model_file):
-            raise RuntimeError(f"{model_file} does not exist")
-        state_dict = torch.load(model_file, map_location="cpu")
-        for k, v in model.state_dict().items():
-            if '_temp.' in k:
-                state_dict.update({k: v})
-        model.load_state_dict(state_dict)
-
-        return model

Tune-A-Video/tuneavideo/models/unet_blocks.py

@@ -1,588 +0,0 @@
-# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py
-
-import torch
-from torch import nn
-
-from .attention import Transformer3DModel
-from .resnet import Downsample3D, ResnetBlock3D, Upsample3D
-
-
-def get_down_block(
-    down_block_type,
-    num_layers,
-    in_channels,
-    out_channels,
-    temb_channels,
-    add_downsample,
-    resnet_eps,
-    resnet_act_fn,
-    attn_num_head_channels,
-    resnet_groups=None,
-    cross_attention_dim=None,
-    downsample_padding=None,
-    dual_cross_attention=False,
-    use_linear_projection=False,
-    only_cross_attention=False,
-    upcast_attention=False,
-    resnet_time_scale_shift="default",
-):
-    down_block_type = down_block_type[7:] if down_block_type.startswith("UNetRes") else down_block_type
-    if down_block_type == "DownBlock3D":
-        return DownBlock3D(
-            num_layers=num_layers,
-            in_channels=in_channels,
-            out_channels=out_channels,
-            temb_channels=temb_channels,
-            add_downsample=add_downsample,
-            resnet_eps=resnet_eps,
-            resnet_act_fn=resnet_act_fn,
-            resnet_groups=resnet_groups,
-            downsample_padding=downsample_padding,
-            resnet_time_scale_shift=resnet_time_scale_shift,
-        )
-    elif down_block_type == "CrossAttnDownBlock3D":
-        if cross_attention_dim is None:
-            raise ValueError("cross_attention_dim must be specified for CrossAttnDownBlock3D")
-        return CrossAttnDownBlock3D(
-            num_layers=num_layers,
-            in_channels=in_channels,
-            out_channels=out_channels,
-            temb_channels=temb_channels,
-            add_downsample=add_downsample,
-            resnet_eps=resnet_eps,
-            resnet_act_fn=resnet_act_fn,
-            resnet_groups=resnet_groups,
-            downsample_padding=downsample_padding,
-            cross_attention_dim=cross_attention_dim,
-            attn_num_head_channels=attn_num_head_channels,
-            dual_cross_attention=dual_cross_attention,
-            use_linear_projection=use_linear_projection,
-            only_cross_attention=only_cross_attention,
-            upcast_attention=upcast_attention,
-            resnet_time_scale_shift=resnet_time_scale_shift,
-        )
-    raise ValueError(f"{down_block_type} does not exist.")
-
-
-def get_up_block(
-    up_block_type,
-    num_layers,
-    in_channels,
-    out_channels,
-    prev_output_channel,
-    temb_channels,
-    add_upsample,
-    resnet_eps,
-    resnet_act_fn,
-    attn_num_head_channels,
-    resnet_groups=None,
-    cross_attention_dim=None,
-    dual_cross_attention=False,
-    use_linear_projection=False,
-    only_cross_attention=False,
-    upcast_attention=False,
-    resnet_time_scale_shift="default",
-):
-    up_block_type = up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type
-    if up_block_type == "UpBlock3D":
-        return UpBlock3D(
-            num_layers=num_layers,
-            in_channels=in_channels,
-            out_channels=out_channels,
-            prev_output_channel=prev_output_channel,
-            temb_channels=temb_channels,
-            add_upsample=add_upsample,
-            resnet_eps=resnet_eps,
-            resnet_act_fn=resnet_act_fn,
-            resnet_groups=resnet_groups,
-            resnet_time_scale_shift=resnet_time_scale_shift,
-        )
-    elif up_block_type == "CrossAttnUpBlock3D":
-        if cross_attention_dim is None:
-            raise ValueError("cross_attention_dim must be specified for CrossAttnUpBlock3D")
-        return CrossAttnUpBlock3D(
-            num_layers=num_layers,
-            in_channels=in_channels,
-            out_channels=out_channels,
-            prev_output_channel=prev_output_channel,
-            temb_channels=temb_channels,
-            add_upsample=add_upsample,
-            resnet_eps=resnet_eps,
-            resnet_act_fn=resnet_act_fn,
-            resnet_groups=resnet_groups,
-            cross_attention_dim=cross_attention_dim,
-            attn_num_head_channels=attn_num_head_channels,
-            dual_cross_attention=dual_cross_attention,
-            use_linear_projection=use_linear_projection,
-            only_cross_attention=only_cross_attention,
-            upcast_attention=upcast_attention,
-            resnet_time_scale_shift=resnet_time_scale_shift,
-        )
-    raise ValueError(f"{up_block_type} does not exist.")
-
-
-class UNetMidBlock3DCrossAttn(nn.Module):
-    def __init__(
-        self,
-        in_channels: int,
-        temb_channels: int,
-        dropout: float = 0.0,
-        num_layers: int = 1,
-        resnet_eps: float = 1e-6,
-        resnet_time_scale_shift: str = "default",
-        resnet_act_fn: str = "swish",
-        resnet_groups: int = 32,
-        resnet_pre_norm: bool = True,
-        attn_num_head_channels=1,
-        output_scale_factor=1.0,
-        cross_attention_dim=1280,
-        dual_cross_attention=False,
-        use_linear_projection=False,
-        upcast_attention=False,
-    ):
-        super().__init__()
-
-        self.has_cross_attention = True
-        self.attn_num_head_channels = attn_num_head_channels
-        resnet_groups = resnet_groups if resnet_groups is not None else min(in_channels // 4, 32)
-
-        # there is always at least one resnet
-        resnets = [
-            ResnetBlock3D(
-                in_channels=in_channels,
-                out_channels=in_channels,
-                temb_channels=temb_channels,
-                eps=resnet_eps,
-                groups=resnet_groups,
-                dropout=dropout,
-                time_embedding_norm=resnet_time_scale_shift,
-                non_linearity=resnet_act_fn,
-                output_scale_factor=output_scale_factor,
-                pre_norm=resnet_pre_norm,
-            )
-        ]
-        attentions = []
-
-        for _ in range(num_layers):
-            if dual_cross_attention:
-                raise NotImplementedError
-            attentions.append(
-                Transformer3DModel(
-                    attn_num_head_channels,
-                    in_channels // attn_num_head_channels,
-                    in_channels=in_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
-                    use_linear_projection=use_linear_projection,
-                    upcast_attention=upcast_attention,
-                )
-            )
-            resnets.append(
-                ResnetBlock3D(
-                    in_channels=in_channels,
-                    out_channels=in_channels,
-                    temb_channels=temb_channels,
-                    eps=resnet_eps,
-                    groups=resnet_groups,
-                    dropout=dropout,
-                    time_embedding_norm=resnet_time_scale_shift,
-                    non_linearity=resnet_act_fn,
-                    output_scale_factor=output_scale_factor,
-                    pre_norm=resnet_pre_norm,
-                )
-            )
-
-        self.attentions = nn.ModuleList(attentions)
-        self.resnets = nn.ModuleList(resnets)
-
-    def forward(self, hidden_states, temb=None, encoder_hidden_states=None, attention_mask=None):
-        hidden_states = self.resnets[0](hidden_states, temb)
-        for attn, resnet in zip(self.attentions, self.resnets[1:]):
-            hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states).sample
-            hidden_states = resnet(hidden_states, temb)
-
-        return hidden_states
-
-
-class CrossAttnDownBlock3D(nn.Module):
-    def __init__(
-        self,
-        in_channels: int,
-        out_channels: int,
-        temb_channels: int,
-        dropout: float = 0.0,
-        num_layers: int = 1,
-        resnet_eps: float = 1e-6,
-        resnet_time_scale_shift: str = "default",
-        resnet_act_fn: str = "swish",
-        resnet_groups: int = 32,
-        resnet_pre_norm: bool = True,
-        attn_num_head_channels=1,
-        cross_attention_dim=1280,
-        output_scale_factor=1.0,
-        downsample_padding=1,
-        add_downsample=True,
-        dual_cross_attention=False,
-        use_linear_projection=False,
-        only_cross_attention=False,
-        upcast_attention=False,
-    ):
-        super().__init__()
-        resnets = []
-        attentions = []
-
-        self.has_cross_attention = True
-        self.attn_num_head_channels = attn_num_head_channels
-
-        for i in range(num_layers):
-            in_channels = in_channels if i == 0 else out_channels
-            resnets.append(
-                ResnetBlock3D(
-                    in_channels=in_channels,
-                    out_channels=out_channels,
-                    temb_channels=temb_channels,
-                    eps=resnet_eps,
-                    groups=resnet_groups,
-                    dropout=dropout,
-                    time_embedding_norm=resnet_time_scale_shift,
-                    non_linearity=resnet_act_fn,
-                    output_scale_factor=output_scale_factor,
-                    pre_norm=resnet_pre_norm,
-                )
-            )
-            if dual_cross_attention:
-                raise NotImplementedError
-            attentions.append(
-                Transformer3DModel(
-                    attn_num_head_channels,
-                    out_channels // attn_num_head_channels,
-                    in_channels=out_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
-                    use_linear_projection=use_linear_projection,
-                    only_cross_attention=only_cross_attention,
-                    upcast_attention=upcast_attention,
-                )
-            )
-        self.attentions = nn.ModuleList(attentions)
-        self.resnets = nn.ModuleList(resnets)
-
-        if add_downsample:
-            self.downsamplers = nn.ModuleList(
-                [
-                    Downsample3D(
-                        out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op"
-                    )
-                ]
-            )
-        else:
-            self.downsamplers = None
-
-        self.gradient_checkpointing = False
-
-    def forward(self, hidden_states, temb=None, encoder_hidden_states=None, attention_mask=None):
-        output_states = ()
-
-        for resnet, attn in zip(self.resnets, self.attentions):
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module, return_dict=None):
-                    def custom_forward(*inputs):
-                        if return_dict is not None:
-                            return module(*inputs, return_dict=return_dict)
-                        else:
-                            return module(*inputs)
-
-                    return custom_forward
-
-                hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
-                hidden_states = torch.utils.checkpoint.checkpoint(
-                    create_custom_forward(attn, return_dict=False),
-                    hidden_states,
-                    encoder_hidden_states,
-                )[0]
-            else:
-                hidden_states = resnet(hidden_states, temb)
-                hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states).sample
-
-            output_states += (hidden_states,)
-
-        if self.downsamplers is not None:
-            for downsampler in self.downsamplers:
-                hidden_states = downsampler(hidden_states)
-
-            output_states += (hidden_states,)
-
-        return hidden_states, output_states
-
-
-class DownBlock3D(nn.Module):
-    def __init__(
-        self,
-        in_channels: int,
-        out_channels: int,
-        temb_channels: int,
-        dropout: float = 0.0,
-        num_layers: int = 1,
-        resnet_eps: float = 1e-6,
-        resnet_time_scale_shift: str = "default",
-        resnet_act_fn: str = "swish",
-        resnet_groups: int = 32,
-        resnet_pre_norm: bool = True,
-        output_scale_factor=1.0,
-        add_downsample=True,
-        downsample_padding=1,
-    ):
-        super().__init__()
-        resnets = []
-
-        for i in range(num_layers):
-            in_channels = in_channels if i == 0 else out_channels
-            resnets.append(
-                ResnetBlock3D(
-                    in_channels=in_channels,
-                    out_channels=out_channels,
-                    temb_channels=temb_channels,
-                    eps=resnet_eps,
-                    groups=resnet_groups,
-                    dropout=dropout,
-                    time_embedding_norm=resnet_time_scale_shift,
-                    non_linearity=resnet_act_fn,
-                    output_scale_factor=output_scale_factor,
-                    pre_norm=resnet_pre_norm,
-                )
-            )
-
-        self.resnets = nn.ModuleList(resnets)
-
-        if add_downsample:
-            self.downsamplers = nn.ModuleList(
-                [
-                    Downsample3D(
-                        out_channels, use_conv=True, out_channels=out_channels, padding=downsample_padding, name="op"
-                    )
-                ]
-            )
-        else:
-            self.downsamplers = None
-
-        self.gradient_checkpointing = False
-
-    def forward(self, hidden_states, temb=None):
-        output_states = ()
-
-        for resnet in self.resnets:
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module):
-                    def custom_forward(*inputs):
-                        return module(*inputs)
-
-                    return custom_forward
-
-                hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
-            else:
-                hidden_states = resnet(hidden_states, temb)
-
-            output_states += (hidden_states,)
-
-        if self.downsamplers is not None:
-            for downsampler in self.downsamplers:
-                hidden_states = downsampler(hidden_states)
-
-            output_states += (hidden_states,)
-
-        return hidden_states, output_states
-
-
-class CrossAttnUpBlock3D(nn.Module):
-    def __init__(
-        self,
-        in_channels: int,
-        out_channels: int,
-        prev_output_channel: int,
-        temb_channels: int,
-        dropout: float = 0.0,
-        num_layers: int = 1,
-        resnet_eps: float = 1e-6,
-        resnet_time_scale_shift: str = "default",
-        resnet_act_fn: str = "swish",
-        resnet_groups: int = 32,
-        resnet_pre_norm: bool = True,
-        attn_num_head_channels=1,
-        cross_attention_dim=1280,
-        output_scale_factor=1.0,
-        add_upsample=True,
-        dual_cross_attention=False,
-        use_linear_projection=False,
-        only_cross_attention=False,
-        upcast_attention=False,
-    ):
-        super().__init__()
-        resnets = []
-        attentions = []
-
-        self.has_cross_attention = True
-        self.attn_num_head_channels = attn_num_head_channels
-
-        for i in range(num_layers):
-            res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
-            resnet_in_channels = prev_output_channel if i == 0 else out_channels
-
-            resnets.append(
-                ResnetBlock3D(
-                    in_channels=resnet_in_channels + res_skip_channels,
-                    out_channels=out_channels,
-                    temb_channels=temb_channels,
-                    eps=resnet_eps,
-                    groups=resnet_groups,
-                    dropout=dropout,
-                    time_embedding_norm=resnet_time_scale_shift,
-                    non_linearity=resnet_act_fn,
-                    output_scale_factor=output_scale_factor,
-                    pre_norm=resnet_pre_norm,
-                )
-            )
-            if dual_cross_attention:
-                raise NotImplementedError
-            attentions.append(
-                Transformer3DModel(
-                    attn_num_head_channels,
-                    out_channels // attn_num_head_channels,
-                    in_channels=out_channels,
-                    num_layers=1,
-                    cross_attention_dim=cross_attention_dim,
-                    norm_num_groups=resnet_groups,
-                    use_linear_projection=use_linear_projection,
-                    only_cross_attention=only_cross_attention,
-                    upcast_attention=upcast_attention,
-                )
-            )
-
-        self.attentions = nn.ModuleList(attentions)
-        self.resnets = nn.ModuleList(resnets)
-
-        if add_upsample:
-            self.upsamplers = nn.ModuleList([Upsample3D(out_channels, use_conv=True, out_channels=out_channels)])
-        else:
-            self.upsamplers = None
-
-        self.gradient_checkpointing = False
-
-    def forward(
-        self,
-        hidden_states,
-        res_hidden_states_tuple,
-        temb=None,
-        encoder_hidden_states=None,
-        upsample_size=None,
-        attention_mask=None,
-    ):
-        for resnet, attn in zip(self.resnets, self.attentions):
-            # pop res hidden states
-            res_hidden_states = res_hidden_states_tuple[-1]
-            res_hidden_states_tuple = res_hidden_states_tuple[:-1]
-            hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
-
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module, return_dict=None):
-                    def custom_forward(*inputs):
-                        if return_dict is not None:
-                            return module(*inputs, return_dict=return_dict)
-                        else:
-                            return module(*inputs)
-
-                    return custom_forward
-
-                hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
-                hidden_states = torch.utils.checkpoint.checkpoint(
-                    create_custom_forward(attn, return_dict=False),
-                    hidden_states,
-                    encoder_hidden_states,
-                )[0]
-            else:
-                hidden_states = resnet(hidden_states, temb)
-                hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states).sample
-
-        if self.upsamplers is not None:
-            for upsampler in self.upsamplers:
-                hidden_states = upsampler(hidden_states, upsample_size)
-
-        return hidden_states
-
-
-class UpBlock3D(nn.Module):
-    def __init__(
-        self,
-        in_channels: int,
-        prev_output_channel: int,
-        out_channels: int,
-        temb_channels: int,
-        dropout: float = 0.0,
-        num_layers: int = 1,
-        resnet_eps: float = 1e-6,
-        resnet_time_scale_shift: str = "default",
-        resnet_act_fn: str = "swish",
-        resnet_groups: int = 32,
-        resnet_pre_norm: bool = True,
-        output_scale_factor=1.0,
-        add_upsample=True,
-    ):
-        super().__init__()
-        resnets = []
-
-        for i in range(num_layers):
-            res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
-            resnet_in_channels = prev_output_channel if i == 0 else out_channels
-
-            resnets.append(
-                ResnetBlock3D(
-                    in_channels=resnet_in_channels + res_skip_channels,
-                    out_channels=out_channels,
-                    temb_channels=temb_channels,
-                    eps=resnet_eps,
-                    groups=resnet_groups,
-                    dropout=dropout,
-                    time_embedding_norm=resnet_time_scale_shift,
-                    non_linearity=resnet_act_fn,
-                    output_scale_factor=output_scale_factor,
-                    pre_norm=resnet_pre_norm,
-                )
-            )
-
-        self.resnets = nn.ModuleList(resnets)
-
-        if add_upsample:
-            self.upsamplers = nn.ModuleList([Upsample3D(out_channels, use_conv=True, out_channels=out_channels)])
-        else:
-            self.upsamplers = None
-
-        self.gradient_checkpointing = False
-
-    def forward(self, hidden_states, res_hidden_states_tuple, temb=None, upsample_size=None):
-        for resnet in self.resnets:
-            # pop res hidden states
-            res_hidden_states = res_hidden_states_tuple[-1]
-            res_hidden_states_tuple = res_hidden_states_tuple[:-1]
-            hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
-
-            if self.training and self.gradient_checkpointing:
-
-                def create_custom_forward(module):
-                    def custom_forward(*inputs):
-                        return module(*inputs)
-
-                    return custom_forward
-
-                hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
-            else:
-                hidden_states = resnet(hidden_states, temb)
-
-        if self.upsamplers is not None:
-            for upsampler in self.upsamplers:
-                hidden_states = upsampler(hidden_states, upsample_size)
-
-        return hidden_states

Tune-A-Video/tuneavideo/pipelines/pipeline_tuneavideo.py

@@ -1,407 +0,0 @@
-# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py
-
-import inspect
-from typing import Callable, List, Optional, Union
-from dataclasses import dataclass
-
-import numpy as np
-import torch
-
-from diffusers.utils import is_accelerate_available
-from packaging import version
-from transformers import CLIPTextModel, CLIPTokenizer
-
-from diffusers.configuration_utils import FrozenDict
-from diffusers.models import AutoencoderKL
-from diffusers.pipeline_utils import DiffusionPipeline
-from diffusers.schedulers import (
-    DDIMScheduler,
-    DPMSolverMultistepScheduler,
-    EulerAncestralDiscreteScheduler,
-    EulerDiscreteScheduler,
-    LMSDiscreteScheduler,
-    PNDMScheduler,
-)
-from diffusers.utils import deprecate, logging, BaseOutput
-
-from einops import rearrange
-
-from ..models.unet import UNet3DConditionModel
-
-
-logger = logging.get_logger(__name__)  # pylint: disable=invalid-name
-
-
-@dataclass
-class TuneAVideoPipelineOutput(BaseOutput):
-    videos: Union[torch.Tensor, np.ndarray]
-
-
-class TuneAVideoPipeline(DiffusionPipeline):
-    _optional_components = []
-
-    def __init__(
-        self,
-        vae: AutoencoderKL,
-        text_encoder: CLIPTextModel,
-        tokenizer: CLIPTokenizer,
-        unet: UNet3DConditionModel,
-        scheduler: Union[
-            DDIMScheduler,
-            PNDMScheduler,
-            LMSDiscreteScheduler,
-            EulerDiscreteScheduler,
-            EulerAncestralDiscreteScheduler,
-            DPMSolverMultistepScheduler,
-        ],
-    ):
-        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)
-
-        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,
-        )
-        self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
-
-    def enable_vae_slicing(self):
-        self.vae.enable_slicing()
-
-    def disable_vae_slicing(self):
-        self.vae.disable_slicing()
-
-    def enable_sequential_cpu_offload(self, gpu_id=0):
-        if is_accelerate_available():
-            from accelerate import cpu_offload
-        else:
-            raise ImportError("Please install accelerate via `pip install accelerate`")
-
-        device = torch.device(f"cuda:{gpu_id}")
-
-        for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
-            if cpu_offloaded_model is not None:
-                cpu_offload(cpu_offloaded_model, device)
-
-
-    @property
-    def _execution_device(self):
-        if self.device != torch.device("meta") or not hasattr(self.unet, "_hf_hook"):
-            return self.device
-        for module in self.unet.modules():
-            if (
-                hasattr(module, "_hf_hook")
-                and hasattr(module._hf_hook, "execution_device")
-                and module._hf_hook.execution_device is not None
-            ):
-                return torch.device(module._hf_hook.execution_device)
-        return self.device
-
-    def _encode_prompt(self, prompt, device, num_videos_per_prompt, do_classifier_free_guidance, negative_prompt):
-        batch_size = len(prompt) if isinstance(prompt, list) else 1
-
-        text_inputs = self.tokenizer(
-            prompt,
-            padding="max_length",
-            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
-
-        text_embeddings = self.text_encoder(
-            text_input_ids.to(device),
-            attention_mask=attention_mask,
-        )
-        text_embeddings = text_embeddings[0]
-
-        # duplicate text embeddings for each generation per prompt, using mps friendly method
-        bs_embed, seq_len, _ = text_embeddings.shape
-        text_embeddings = text_embeddings.repeat(1, num_videos_per_prompt, 1)
-        text_embeddings = text_embeddings.view(bs_embed * num_videos_per_prompt, seq_len, -1)
-
-        # get unconditional embeddings for classifier free guidance
-        if do_classifier_free_guidance:
-            uncond_tokens: List[str]
-            if negative_prompt is None:
-                uncond_tokens = [""] * batch_size
-            elif 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
-
-            max_length = text_input_ids.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
-
-            uncond_embeddings = self.text_encoder(
-                uncond_input.input_ids.to(device),
-                attention_mask=attention_mask,
-            )
-            uncond_embeddings = uncond_embeddings[0]
-
-            # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
-            seq_len = uncond_embeddings.shape[1]
-            uncond_embeddings = uncond_embeddings.repeat(1, num_videos_per_prompt, 1)
-            uncond_embeddings = uncond_embeddings.view(batch_size * num_videos_per_prompt, seq_len, -1)
-
-            # 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
-            text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
-
-        return text_embeddings
-
-    def decode_latents(self, latents):
-        video_length = latents.shape[2]
-        latents = 1 / 0.18215 * latents
-        latents = rearrange(latents, "b c f h w -> (b f) c h w")
-        video = self.vae.decode(latents).sample
-        video = rearrange(video, "(b f) c h w -> b c f h w", f=video_length)
-        video = (video / 2 + 0.5).clamp(0, 1)
-        # we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
-        video = video.cpu().float().numpy()
-        return video
-
-    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):
-        if 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 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 None) or (
-            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)}."
-            )
-
-    def prepare_latents(self, batch_size, num_channels_latents, video_length, height, width, dtype, device, generator, latents=None):
-        shape = (batch_size, num_channels_latents, video_length, height // self.vae_scale_factor, 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:
-            rand_device = "cpu" if device.type == "mps" else device
-
-            if isinstance(generator, list):
-                shape = (1,) + shape[1:]
-                latents = [
-                    torch.randn(shape, generator=generator[i], device=rand_device, dtype=dtype)
-                    for i in range(batch_size)
-                ]
-                latents = torch.cat(latents, dim=0).to(device)
-            else:
-                latents = torch.randn(shape, generator=generator, device=rand_device, dtype=dtype).to(device)
-        else:
-            if latents.shape != shape:
-                raise ValueError(f"Unexpected latents shape, got {latents.shape}, expected {shape}")
-            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
-
-    @torch.no_grad()
-    def __call__(
-        self,
-        prompt: Union[str, List[str]],
-        video_length: Optional[int],
-        height: Optional[int] = None,
-        width: Optional[int] = None,
-        num_inference_steps: int = 50,
-        guidance_scale: float = 7.5,
-        negative_prompt: Optional[Union[str, List[str]]] = None,
-        num_videos_per_prompt: Optional[int] = 1,
-        eta: float = 0.0,
-        generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
-        latents: Optional[torch.FloatTensor] = None,
-        output_type: Optional[str] = "tensor",
-        return_dict: bool = True,
-        callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
-        callback_steps: Optional[int] = 1,
-        **kwargs,
-    ):
-        # Default height and width to unet
-        height = height or self.unet.config.sample_size * self.vae_scale_factor
-        width = width or self.unet.config.sample_size * self.vae_scale_factor
-
-        # Check inputs. Raise error if not correct
-        self.check_inputs(prompt, height, width, callback_steps)
-
-        # Define call parameters
-        batch_size = 1 if isinstance(prompt, str) else len(prompt)
-        device = self._execution_device
-        # 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.
-        do_classifier_free_guidance = guidance_scale > 1.0
-
-        # Encode input prompt
-        text_embeddings = self._encode_prompt(
-            prompt, device, num_videos_per_prompt, do_classifier_free_guidance, negative_prompt
-        )
-
-        # Prepare timesteps
-        self.scheduler.set_timesteps(num_inference_steps, device=device)
-        timesteps = self.scheduler.timesteps
-
-        # Prepare latent variables
-        num_channels_latents = self.unet.in_channels
-        latents = self.prepare_latents(
-            batch_size * num_videos_per_prompt,
-            num_channels_latents,
-            video_length,
-            height,
-            width,
-            text_embeddings.dtype,
-            device,
-            generator,
-            latents,
-        )
-        latents_dtype = latents.dtype
-
-        # Prepare extra step kwargs.
-        extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
-
-        # Denoising loop
-        num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
-        with self.progress_bar(total=num_inference_steps) as progress_bar:
-            for i, t in enumerate(timesteps):
-                # expand the latents if we are doing classifier free guidance
-                latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
-                latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
-
-                # predict the noise residual
-                noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample.to(dtype=latents_dtype)
-
-                # perform guidance
-                if do_classifier_free_guidance:
-                    noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
-                    noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
-
-                # compute the previous noisy sample x_t -> x_t-1
-                latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
-
-                # 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:
-                        callback(i, t, latents)
-
-        # Post-processing
-        video = self.decode_latents(latents)
-
-        # Convert to tensor
-        if output_type == "tensor":
-            video = torch.from_numpy(video)
-
-        if not return_dict:
-            return video
-
-        return TuneAVideoPipelineOutput(videos=video)

Tune-A-Video/tuneavideo/util.py

@@ -1,23 +0,0 @@
-import os
-import imageio
-import numpy as np
-
-import torch
-import torchvision
-
-from einops import rearrange
-
-
-def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=4, fps=3):
-    videos = rearrange(videos, "b c t h w -> t b c h w")
-    outputs = []
-    for x in videos:
-        x = torchvision.utils.make_grid(x, nrow=n_rows)
-        x = x.transpose(0, 1).transpose(1, 2).squeeze(-1)
-        if rescale:
-            x = (x + 1.0) / 2.0  # -1,1 -> 0,1
-        x = (x * 255).numpy().astype(np.uint8)
-        outputs.append(x)
-
-    os.makedirs(os.path.dirname(path), exist_ok=True)
-    imageio.mimsave(path, outputs, fps=fps)

inference_fatezero.py

@@ -16,16 +16,16 @@ def merge_config_then_run(
         enhance_words_value,
         num_steps,
         guidance_scale,
-        user_input_video,
+        user_input_video=None,
         
         # Temporal and spatial crop of the video
-        start_sample_frame,
-        n_sample_frame,
-        stride,
-        left_crop,
-        right_crop,
-        top_crop,
-        bottom_crop,
+        start_sample_frame=0,
+        n_sample_frame=8,
+        stride=1,
+        left_crop=0,
+        right_crop=0,
+        top_crop=0,
+        bottom_crop=0,
     ):
     # , ] = inputs
     default_edit_config='FateZero/config/low_resource_teaser/jeep_watercolor_ddim_10_steps.yaml'

patch

@@ -1,15 +0,0 @@
-diff --git a/train_tuneavideo.py b/train_tuneavideo.py
-index 66d51b2..86b2a5d 100644
---- a/train_tuneavideo.py
-+++ b/train_tuneavideo.py
-@@ -94,8 +94,8 @@ def main(
- 
-     # Handle the output folder creation
-     if accelerator.is_main_process:
--        now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
--        output_dir = os.path.join(output_dir, now)
-+        #now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
-+        #output_dir = os.path.join(output_dir, now)
-         os.makedirs(output_dir, exist_ok=True)
-         OmegaConf.save(config, os.path.join(output_dir, 'config.yaml'))
-