PyVision-RL: Forging Open Agentic Vision Models via RL
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866912923157266432 |
|---|---|
| author | Zhao, Shitian Lin, Shaoheng Li, Ming Zhang, Haoquan Peng, Wenshuo Zhang, Kaipeng Wei, Chen |
| author_facet | Zhao, Shitian Lin, Shaoheng Li, Ming Zhang, Haoquan Peng, Wenshuo Zhang, Kaipeng Wei, Chen |
| contents | Reinforcement learning for agentic multimodal models often suffers from interaction collapse, where models learn to reduce tool usage and multi-turn reasoning, limiting the benefits of agentic behavior. We introduce PyVision-RL, a reinforcement learning framework for open-weight multimodal models that stabilizes training and sustains interaction. Our approach combines an oversampling-filtering-ranking rollout strategy with an accumulative tool reward to prevent collapse and encourage multi-turn tool use. Using a unified training pipeline, we develop PyVision-Image and PyVision-Video for image and video understanding. For video reasoning, PyVision-Video employs on-demand context construction, selectively sampling task-relevant frames during reasoning to significantly reduce visual token usage. Experiments show strong performance and improved efficiency, demonstrating that sustained interaction and on-demand visual processing are critical for scalable multimodal agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_20739 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | PyVision-RL: Forging Open Agentic Vision Models via RL Zhao, Shitian Lin, Shaoheng Li, Ming Zhang, Haoquan Peng, Wenshuo Zhang, Kaipeng Wei, Chen Artificial Intelligence Computer Vision and Pattern Recognition Reinforcement learning for agentic multimodal models often suffers from interaction collapse, where models learn to reduce tool usage and multi-turn reasoning, limiting the benefits of agentic behavior. We introduce PyVision-RL, a reinforcement learning framework for open-weight multimodal models that stabilizes training and sustains interaction. Our approach combines an oversampling-filtering-ranking rollout strategy with an accumulative tool reward to prevent collapse and encourage multi-turn tool use. Using a unified training pipeline, we develop PyVision-Image and PyVision-Video for image and video understanding. For video reasoning, PyVision-Video employs on-demand context construction, selectively sampling task-relevant frames during reasoning to significantly reduce visual token usage. Experiments show strong performance and improved efficiency, demonstrating that sustained interaction and on-demand visual processing are critical for scalable multimodal agents. |
| title | PyVision-RL: Forging Open Agentic Vision Models via RL |
| topic | Artificial Intelligence Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2602.20739 |