VITRON: A Unified Pixel-level Vision LLM for Understanding, Generating, Segmenting, Editing

Hao Fei$^{1,2}$, Shengqiong Wu$^{1,2}$, Hanwang Zhang$^{1,3}$, Tat-Seng Chua$^{2}$, Shuicheng Yan$^{1}$

▶ $^{1}$ Skywork AI, Singapore ▶ $^{2}$ National University of Singapore ▶ $^{3}$ Nanyang Technological University

📰 News

[2024.04.04] 👀👀👀 Our Vitron is available now! Welcome to watch 👀 this repository for the latest updates.

😮 Highlights

Existing vision LLMs might still encounter challenges such as superficial instance-level understanding, lack of unified support for both images and videos, and insufficient coverage across various vision tasks. To fill the gaps, we present Vitron, a universal pixel-level vision LLM, designed for comprehensive understanding (perceiving and reasoning), generating, segmenting (grounding and tracking), editing (inpainting) of both static image and dynamic video content.

🛠️ Requirements and Installation

Python >= 3.8
Pytorch == 2.1.0
CUDA Version >= 11.8
Install required packages:

git clone https://github.com/SkyworkAI/Vitron
cd Vitron
conda create -n vitron python=3.10 -y
conda activate vitron
pip install --upgrade pip 
pip install -e .
pip install -e ".[train]"
pip install flash-attn --no-build-isolation
pip install decord opencv-python git+https://github.com/facebookresearch/pytorchvideo.git@28fe037d212663c6a24f373b94cc5d478c8c1a1d

🔥🔥🔥 Installation or Running Fails? 🔥🔥🔥

When running ffmpeg, Unknown encoder 'x264':

try to re-install ffmpeg:

conda uninstall ffmpeg
conda install -c conda-forge ffmpeg   # `-c conda-forge` can not omit

Fail to install detectron2, try this command:

python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'

or refer this Website.

Error in gradio. As there are a big update in gradio>=4.0.0, please make sure install gradio with the same verion in requirements.txt.

Error with deepspeed. If you fine-tune your model, this error occours:

FAILED: cpu_adam.so
/usr/bin/ld: cannot find -lcurand

This error is caused by the wrong soft links when installing deepspeed. Please try to the following command to solve the error:

cd ~/miniconda3/envs/vitron/lib
ls -al libcurand*  # check the links
rm libcurand.so   # remove the wrong links
ln -s libcurand.so.10.3.5.119 libcurand.so  # build new links

Double check again:

python 
from deepspeed.ops.op_builder import CPUAdamBuilder
ds_opt_adam = CPUAdamBuilder().load()  # if loading successfully, then deepspeed are installed successfully.

👍 Deploying Gradio Demo

Firstly, you need to prepare the checkpoint, and then you can run the demo locally via:

python app.py

🙌 Related Projects

You may refer to related work that serves as foundations for our framework and code repository, Vicuna, SEEM, i2vgenxl, StableVideo, and Zeroscope. We also partially draw inspirations from Video-LLaVA, and LanguageBind. Thanks for their wonderful works.

🔒 License

The majority of this project is released under the Apache 2.0 license as found in the LICENSE file.
The service is a research preview intended for non-commercial use only, subject to the model License of LLaMA, Terms of Use of the data generated by OpenAI, and Privacy Practices of ShareGPT. Please contact us if you find any potential violation.

✏️ Citation

If you find our paper and code useful in your research, please consider giving a star ⭐ and citation 📝.

@articles{hao2024vitron,
  title={Vitron: A Unified Pixel-level Vision LLM for Understanding, Generating, Segmenting, Editing},
  author={Hao Fei, Shengqiong Wu, Hanwang Zhang, Tat-Seng Chua, Shuicheng Yan},
  journal={CoRR},
  year={2024}
}

SkyworkAI / Vitron