Open-Sora Plan

[Project Page] [中文主页]

     

💪 Goal

This project aims to create a simple and scalable repo, to reproduce Sora (OpenAI, but we prefer to call it "ClosedAI" ) and build knowledge about Video-VQVAE (VideoGPT) + DiT at scale. However, since we have limited resources, we deeply wish all open-source community can contribute to this project. Pull requests are welcome!!!

本项目希望通过开源社区的力量复现Sora，由北大-兔展AIGC联合实验室共同发起，当前我们资源有限仅搭建了基础架构，无法进行完整训练，希望通过开源社区逐步增加模块并筹集资源进行训练，当前版本离目标差距巨大，仍需持续完善和快速迭代，欢迎Pull request！！！

Project stages:

• Primary

1. Setup the codebase and train a un-conditional model on a landscape dataset.
2. Train models that boost resolution and duration.

• Extensions

3. Conduct text2video experiments on landscape dataset.
4. Train the 1080p model on video2text dataset.
5. Control model with more conditions.

📰 News

[2024.03.27] 🚀🚀🚀 We release the report of VideoCausalVAE, which supports both images and videos. We present our reconstructed video in this demonstration as follows. The text-to-video model is on the way.

test_ducks.mp4

[2024.03.10] 🚀🚀🚀 This repo supports training a latent size of 225×90×90 (t×h×w), which means we are able to train 1 minute of 1080P video with 30FPS (2× interpolated frames and 2× super resolution) under class-condition.

[2024.03.08] We support the training code of text condition with 16 frames of 512x512. The code is mainly borrowed from Latte.

[2024.03.07] We support training with 128 frames (when sample rate = 3, which is about 13 seconds) of 256x256, or 64 frames (which is about 6 seconds) of 512x512.

[2024.03.05] See our latest todo, pull requests are welcome.

[2024.03.04] We re-organizes and modulizes our code to make it easy to contribute to the project, to contribute please see the Repo structure.

[2024.03.03] We opened some discussions to clarify several issues.

[2024.03.01] Training code is available now! Learn more on our project page. Please feel free to watch 👀 this repository for the latest updates.

✊ Todo

Setup the codebase and train a unconditional model on landscape dataset

• Fix typos & Update readme. 🤝 Thanks to @mio2333, @CreamyLong, @chg0901, @Nyx-177, @HowardLi1984, @sennnnn, @Jason-fan20
• Setup environment. 🤝 Thanks to @nameless1117
• Add docker file. ⌛ [WIP] 🤝 Thanks to @Mon-ius, @SimonLeeGit
• Enable type hints for functions. 🤝 Thanks to @RuslanPeresy, 🙏 [Need your contribution]
• Resume from checkpoint.
• Add Video-VQGAN model, which is borrowed from VideoGPT.
• Support variable aspect ratios, resolutions, durations training on DiT.
• Support Dynamic mask input inspired by FiT.
• Add class-conditioning on embeddings.
• Incorporating Latte as main codebase.
• Add VAE model, which is borrowed from Stable Diffusion.
• Joint dynamic mask input with VAE.
• Add VQVAE from VQGAN. 🙏 [Need your contribution]
• Make the codebase ready for the cluster training. Add SLURM scripts. 🙏 [Need your contribution]
• Refactor VideoGPT. 🤝 Thanks to @qqingzheng, @luo3300612, @sennnnn
• Add sampling script.
• Add DDP sampling script. ⌛ [WIP]
• Use accelerate on multi-node. 🤝 Thanks to @sysuyy
• Incorporate SiT. 🤝 Thanks to @khan-yin
• Add evaluation scripts (FVD, CLIP score). 🤝 Thanks to @rain305f

Train models that boost resolution and duration

• Add PI to support out-of-domain size. 🤝 Thanks to @jpthu17
• Add 2D RoPE to improve generalization ability as FiT. 🤝 Thanks to @jpthu17
• Compress KV according to PixArt-sigma.
• Support deepspeed for videogpt training. 🤝 Thanks to @sennnnn
• Train a low dimension Video-AE, whether it is VAE or VQVAE. ⌛ [WIP] 🚀 [Require more computation]
• Extract offline feature.
• Train with offline feature.
• Add frame interpolation model. 🤝 Thanks to @yunyangge
• Add super resolution model. 🤝 Thanks to @Linzy19
• Add accelerate to automatically manage training.
• Joint training with images. 🙏 [Need your contribution]
• Implement MaskDiT technique for fast training. 🙏 [Need your contribution]
• Incorporate NaViT. 🙏 [Need your contribution]
• Add FreeNoise support for training-free longer video generation. 🙏 [Need your contribution]

Conduct text2video experiments on landscape dataset.

• Implement PeRFlow for improving the sampling process. 🙏 [Need your contribution]
• Finish data loading, pre-processing utils.
• Add T5 support.
• Add CLIP support. 🤝 Thanks to @Ytimed2020
• Add text2image training script.
• Add prompt captioner.
  • Collect training data.
    • Need video-text pairs with caption. 🙏 [Need your contribution]
    • Extract multi-frame descriptions by large image-language models. 🤝 Thanks to @HowardLi1984
    • Extract video description by large video-language models. 🙏 [Need your contribution]
    • Integrate captions to get a dense caption by using a large language model, such as GPT-4. 🤝 Thanks to @HowardLi1984
  • Train a captioner to refine captions. 🚀 [Require more computation]

Train the 1080p model on video2text dataset

• Looking for a suitable dataset, welcome to discuss and recommend. 🙏 [Need your contribution]
• Add synthetic video created by game engines or 3D representations. 🙏 [Need your contribution]
• Finish data loading, and pre-processing utils. ⌛ [WIP]
• Support memory friendly training.
  • Add flash-attention2 from pytorch.
  • Add xformers. 🤝 Thanks to @jialin-zhao
  • Support mixed precision training.
  • Add gradient checkpoint.
  • Support for ReBased and Ring attention. 🤝 Thanks to @kabachuha
  • Train using the deepspeed engine. 🤝 Thanks to @sennnnn
  • Integrate with Colossal-AI for a cheaper, faster, and more efficient. 🙏 [Need your contribution]
• Train with a text condition. Here we could conduct different experiments: 🚀 [Require more computation]
  • Train with T5 conditioning.
  • Train with CLIP conditioning.
  • Train with CLIP + T5 conditioning (probably costly during training and experiments).

Control model with more condition

• Load pretrained weights from Latte.
• Incorporating ControlNet. 🙏 [Need your contribution]

📂 Repo structure (WIP)

├── README.md
├── docs
│   ├── Data.md                    -> Datasets description.
│   ├── Contribution_Guidelines.md -> Contribution guidelines description.
├── scripts                        -> All scripts.
├── opensora
│   ├── dataset
│   ├── models
│   │   ├── ae                     -> Compress videos to latents
│   │   │   ├── imagebase
│   │   │   │   ├── vae
│   │   │   │   └── vqvae
│   │   │   └── videobase
│   │   │       ├── vae
│   │   │       └── vqvae
│   │   ├── captioner
│   │   ├── diffusion              -> Denoise latents
│   │   │   ├── diffusion         
│   │   │   ├── dit
│   │   │   ├── latte
│   │   │   └── unet
│   │   ├── frame_interpolation
│   │   ├── super_resolution
│   │   └── text_encoder
│   ├── sample
│   ├── train                      -> Training code
│   └── utils

🛠️ Requirements and Installation

1. Clone this repository and navigate to Open-Sora-Plan folder

git clone https://github.com/PKU-YuanGroup/Open-Sora-Plan
cd Open-Sora-Plan

2. Install required packages

conda create -n opensora python=3.8 -y
conda activate opensora
pip install -e .

3. Install additional packages for training cases

pip install -e ".[train]"
pip install flash-attn --no-build-isolation

4. Install optional requirements such as static type checking:

pip install -e '.[dev]'

🗝️ Usage

Datasets

Refer to Data.md

Evaluation

Refer to the document EVAL.md.

Causal Video VAE

Reconstructing

python examples/rec_video_vae.py --rec-path test_video.mp4 --video-path video.mp4 --resolution 512 --num-frames 1440 --sample-rate 1 --sample-fps 24 -
-device cuda --ckpt <Your ckpt>

For more details, please refer to: CausalVideoVAE Report.

VideoGPT VQVAE

Please refer to the document VQVAE.

Video Diffusion Transformer

Training

sh scripts/train.sh

The current resources are only enough for us to do primary experiments on the Sky dataset.

Sampling

sh scripts/sample.sh

Below is a visualization of the sampling results.

12s 256x256	25s 256x256

🚀 Improved Training Performance

In comparison to the original implementation, we implement a selection of training speed acceleration and memory saving features including gradient checkpointing, mixed precision training, and pre-extracted features, xformers, deepspeed. Some data points using a batch size of 1 with a A100:

64×32×32 (origin size: 256×256×256)

gradient checkpointing	mixed precision	xformers	feature pre-extraction	deepspeed config	compress kv	training speed	memory
✔	✔	✔	✔	❌	❌	0.64 steps/sec	43G
✔	✔	✔	✔	Zero2	❌	0.66 steps/sec	14G
✔	✔	✔	✔	Zero2	✔	0.66 steps/sec	15G
✔	✔	✔	✔	Zero2 offload	❌	0.33 steps/sec	11G
✔	✔	✔	✔	Zero2 offload	✔	0.31 steps/sec	12G

128×64×64 (origin size: 512×512×512)

gradient checkpointing	mixed precision	xformers	feature pre-extraction	deepspeed config	compress kv	training speed	memory
✔	✔	✔	✔	❌	❌	0.08 steps/sec	77G
✔	✔	✔	✔	Zero2	❌	0.08 steps/sec	41G
✔	✔	✔	✔	Zero2	✔	0.09 steps/sec	36G
✔	✔	✔	✔	Zero2 offload	❌	0.07 steps/sec	39G
✔	✔	✔	✔	Zero2 offload	✔	0.07 steps/sec	33G

💡 How to Contribute to the Open-Sora Plan Community

We greatly appreciate your contributions to the Open-Sora Plan open-source community and helping us make it even better than it is now!

For more details, please refer to the Contribution Guidelines

👍 Acknowledgement

• Latte: The main codebase we built upon and it is an wonderful video gererated model.
• VideoGPT: Video Generation using VQ-VAE and Transformers.
• DiT: Scalable Diffusion Models with Transformers.
• FiT: Flexible Vision Transformer for Diffusion Model.
• Positional Interpolation: Extending Context Window of Large Language Models via Positional Interpolation.

🔒 License

• See LICENSE for details.

🤝 Community contributors