base_model:
- OpenGVLab/InternVideo2_5_Chat_8B
datasets:
- yzy666/SVBench
language:
- en
license: apache-2.0
metrics:
- code_eval
pipeline_tag: video-text-to-text
library_name: transformers
Model Card for StreamingChat
This model card aims to provide a comprehensive overview of the StreamingChat model. For details, see our Project, Paper, Dataset and GitHub repository.
Dataset Description
StreamingChat is a streaming video understanding model built upon InternVideo2.5. It utilizes Streaming video dialogue data, including temporal dialogue paths from the SVBench training set. The model is fine-tuned using a static resolution strategy, enabling it to process several minutes of video at a rate of 1 FPS. Images are interleaved with language tokens, with each image comprising 16 tokens. This model aims to catalyze progress in streaming video understanding.
Uses
Download the StreamingChat model from Hugging Face:
git clone https://huggingface.co/yzy666/StreamingChat_8B
Install Python dependencies:
conda create -n StreamingChat -y python=3.9.21
conda activate StreamingChat
conda install -y -c pytorch pytorch=2.5.1 torchvision=0.10.1
pip install transformers=4.37.2 opencv-python=4.11.0.84 imageio=2.37.0 decord=0.6.0
pip install flash-attn --no-build-isolation
Run the inference script directly：
python demo.py
Citation
If you find our data useful, please consider citing our work!
@article{yang2025svbench,
title={SVBench: A Benchmark with Temporal Multi-Turn Dialogues for Streaming Video Understanding},
author={Yang, Zhenyu and Hu, Yuhang and Du, Zemin and Xue, Dizhan and Qian, Shengsheng and Wu, Jiahong and Yang, Fan and Dong, Weiming and Xu, Changsheng},
journal={arXiv preprint arXiv:2502.10810},
year={2025}
}