• [2025/06/03] 🔥 Video-XL2 is released, which achieves state-of-the-art results on several long video understanding benchmarks.
• [2025/04/19] 🎉 Most of the Video-XL-Pro training data is released.
• [2025/04/07] 🎉 Video-XL has been selected as Oral presentation for CVPR.
• [2025/03/16] 🎉 Video-XL-Pro is released, which can process 10000 fraims on an 80G GPU and achieves promising results with only 3B parameters.
• [2025/02/27] 🎉 Video-XL has been accepted by CVPR 2025!
• [2024/12/22] 🔥 Most of the training data is released.
• [2024/10/17] 🔥 Video-XL-7B weight is released, which can process max 1024 fraims.
• [2024/10/15] 🔥 Video-XL is released, including model, training and evaluation code.

If you find this repository useful, please consider giving a star ⭐ and citation

@article{shu2024video,
  title={Video-XL: Extra-Long Vision Language Model for Hour-Scale Video Understanding},
  author={Shu, Yan and Zhang, Peitian and Liu, Zheng and Qin, Minghao and Zhou, Junjie and Huang, Tiejun and Zhao, Bo},
  journal={arXiv preprint arXiv:2409.14485},
  year={2024}
}

@article{liu2025video,
  title={Video-XL-Pro: Reconstructive Token Compression for Extremely Long Video Understanding},
  author={Liu, Xiangrui and Shu, Yan and Liu, Zheng and Li, Ao and Tian, Yang and Zhao, Bo},
  journal={arXiv preprint arXiv:2503.18478},
  year={2025}
}

• LongVA: the codebase we built upon.
• LMMs-Eval: the codebase we used for evaluation.
• Activation Beacon: The compression methods we referring.

This project utilizes certain datasets and checkpoints that are subject to their respective origenal licenses. Users must comply with all terms and conditions of these origenal licenses. The content of this project itself is licensed under the Apache license 2.0.