README.md

---
license: mit
---

[Robust-R1:Degradation-Aware Reasoning for Robust Visual Understanding](https://arxiv.org/abs/2512.17532)

## 🏰 **Pretrained and Fine-tuned Model**

- The following checkpoints are utilized to run Robust-R1:

  | Checkpoint | Link | Note |
  |:---------:|:----:|:----:|
  | Qwen2.5-VL-Base | [link](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct) | Used as initial weights for training. |
  | **Robust-R1-SFT** | [link](https://huggingface.co/Jiaqi-hkust/Robust-R1-SFT) | Fine-tuned on [Robust-R1 dataset](https://huggingface.co/datasets/Jiaqi-hkust/Robust-R1) |
  | **Robust-R1-RL** | [link](https://huggingface.co/Jiaqi-hkust/Robust-R1-RL) | Fine-tuned with reinforcement learning on [Robust-R1 dataset](https://huggingface.co/datasets/Jiaqi-hkust/Robust-R1) |

## ⭐️ Citation

If you find Robust-R1 useful for your research and applications, please cite using this BibTeX:
   ``` latex
   @inproceedings{tang2025robustr1,
     title={Robust-R1: Degradation-Aware Reasoning for Robust Visual Understanding},
     author={Tang, Jiaqi and Chen, Jianmin and Wei, Wei and Xu, Xiaogang and Liu, Runtao and Wu, Xiangyu and Xie, Qipeng and Wu, Jiafei and Zhang, Lei and Chen, Qifeng},
     booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
     year={2026}
   }
   ```