Research map / Vision-language and multimodal / Multimodal LLMs
Visual-RFT: Visual Reinforcement Fine-Tuning
Ziyu Liu, Zeyi Sun, Yuhang Zang, Xiaoyi Dong, Yuhang Cao, Haodong Duan, Dahua Lin, Jiaqi Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace SFT with Visual-RFT (GRPO with verifiable rewards) | Qwen2-VL-2B · Fine-grained classification one-shot | accuracy: +24.3 | Improved |
| use SFT on one-shot classification | Qwen2-VL-2B · Fine-grained classification one-shot | accuracy: -4.3 | Hurt |
| replace SFT with Visual-RFT | Qwen2-VL-2B · Fine-grained classification 16-shot | accuracy: 64.0 -> 85.3 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-2B · COCO 8 categories 2-shot | mAP: 21.0 -> 41.5 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-2B · COCO 8 categories 16-shot | mAP: 31.3 -> 46.8 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-7B · COCO 8 categories 4-shot | mAP: 44.1 -> 54.3 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-2B · LVIS 6 rare categories 10-shot | mAP: 10.0 -> 19.4 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-2B · MG 5 categories 4-shot (out-of-domain) | mAP: 26.8 -> 61.8 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-2B · LISA reasoning grounding | mIoU test: 28.3 -> 37.6 | Improved |
| replace SFT with Visual-RFT | Qwen2-VL-7B · LISA reasoning grounding | mIoU test: 39.1 -> 43.9 | Improved |
| use SFT on 7B for COCO novel classes | Qwen2-VL-7B · COCO open-vocab novel | mAP novel: 26.3 -> 25.7 | No clear effect |
| replace SFT with Visual-RFT | Qwen2-VL-7B · COCO open-vocab novel | mAP novel: 25.7 -> 35.8 | Improved |
Later papers that compared with this method
4 later accepted papers report Visual-RFT as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Dual-Latent Memory Routing for Vision-Language Reasoning (ICML 2026 · better on 7 of 7 shared results)
- Start Small, Think Big: Curriculum-based Relative Policy Optimization for Visual Grounding (AAAI 2026 · better on 7 of 7 shared results)
- VisMem: Latent Vision Memory Unlocks Potential of Vision-Language Models (CVPR 2026 · better on 1 of 1 shared result)
- SAM-R1: Leveraging SAM for Reward Feedback in Multimodal Segmentation via Reinforcement Learning (NeurIPS 2025 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models (NeurIPS 2025 · similarity 0.93)
- RewardMap: Tackling Sparse Rewards in Fine-grained Visual Reasoning via Multi-Stage Reinforcement Learning (ICLR 2026 · similarity 0.89)
- VideoRFT: Incentivizing Video Reasoning Capability in MLLMs via Reinforced Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- To Think or Not To Think: A Study of Thinking in Rule-Based Visual Reinforcement Fine-Tuning (NeurIPS 2025 · similarity 0.89)
- Fine-Tuning Large Vision-Language Models as Decision-Making Agents via Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
