VideoPro: Adaptive Program Reasoning for Long Video Understanding

Chenglin Li1,4, Feng Han2,4, Yikun Wang2,4, Ruilin Li3,4, Shuai Dong4,
Haowen Hou4, Haitao Li1,2, Qianglong Chen1, FengTao1, Jingqi Tong2,4,
Yin Zhang1, Jiaqi Wang4
1Zhejiang University, 2Fudan University, 3Wuhan University,
4Shanghai Innovation Institute
Paper PDF Hugging Face Spaces

VideoPro Overview

pipeline
data-overview

Qualitative Cases of Adaptive Program Reasoning

pipeline

Benchmarking Results on Long-video benchmarks

data-overview

Performance and Efficiency Comparison.

data-overview

Accuracy on LongVideoBench and VideoMME across Video Durations.

pipeline

BibTeX

@article{li2025videopro,
  title={VideoPro: Adaptive Program Reasoning for Long Video Understanding},
  author={Li, Chenglin and Han, Feng and Wang, Yikun and Li, Ruilin and Dong, Shuai and Hou, Haowen and Li, Haitao and Chen, Qianglong and Tao, Feng and Tong, Jingqi and others},
  journal={arXiv preprint arXiv:2509.17743},
  year={2025}
}