ByteDance/Q-Insight

Model Description

This is a huggingface model card for Q-Insight family 👋

• Q-Insight paper: https://arxiv.org/pdf/2503.22679
• VQ-Insight paper: https://arxiv.org/pdf/2506.18564
• RALI paper: https://arxiv.org/pdf/2510.11369
• Code: https://github.com/bytedance/Q-Insight

License

This project is licensed under the Apache-2.0 License. It is finetuned from Qwen2.5-VL-7B-Instruct under Apache-2.0.

Citation

If you find the code helpful in your research or work, please cite the following papers:

@inproceedings{li2025qinsight,
  title={Q-insight: Understanding image quality via visual reinforcement learning},
  author={Li, Weiqi and Zhang, Xuanyu and Zhao, Shijie and Zhang, Yabin and Li, Junlin and Zhang, Li and Zhang, Jian},
  booktitle={Advances in Neural Information Processing Systems},
  year={2025}
}

@inproceedings{zhang2025vqinsight,
  title={VQ-Insight: Teaching VLMs for AI-Generated Video Quality Understanding via Progressive Visual Reinforcement Learning},
  author={Zhang, Xuanyu and Li, Weiqi and Zhao, Shijie and Li, Junlin and Zhang, Li and Zhang, Jian},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  year={2026}
}

@inproceedings{zhao2025reasoning,
  title={Reasoning as Representation: Rethinking Visual Reinforcement Learning in Image Quality Assessment},
  author={Zhao, Shijie and Zhang, Xuanyu and Li, Weiqi and Li, Junlin and Zhang, Li and Xue, Tianfan and Zhang, Jian},
  booktitle={Proceedings of the International Conference on Learning Representations (ICLR)},
  year={2026}
}