Research map / Vision-language and multimodal / Multimodal LLMs
From Illusion to Intention: Visual Rationale Learning for Reliable Evidence Acquisition
Changpeng Wang, Liu Junhan, Xi Chen, Haozhe Wang, Donglian Qi, Yunfeng Yan
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 |
|---|---|---|---|
| add rationale fidelity reward instead of answer-only reward (ablation) | ViRL · V* | answer accuracy: 87.6 -> 90.1 | Improved |
| add rationale fidelity reward instead of answer + tool-call bonus (ablation) | ViRL · V* | answer accuracy: 85.3 -> 90.1 | Improved |
| add fine-grained credit assignment (ablation) | ViRL · V* | answer accuracy: 88.9 -> 90.1 | Improved |
| add fine-grained credit assignment (ablation) | ViRL · V* | rationale accuracy: 78.2 -> 87.4 | Improved |
| use signed plus refinement reward instead of raw IoU reward (ablation) | ViRL · V* | answer accuracy: 84.3 -> 89.8 | Improved |
| use signed plus refinement reward instead of raw IoU reward (ablation) | ViRL · V* | rationale accuracy: 50.8 -> 83.7 | Improved |
| raise soft threshold h0 from 0.2 to 0.3 (ablation) | ViRL · V* | rationale accuracy: 75.5 -> 77.3 | Improved |
| raise soft threshold h0 to 0.4 or higher (ablation) | ViRL · V* | answer accuracy: 88.4 -> 86.6 | Hurt |
| replace Visual-CoT data with evidence-seeking QA (ablation) | ViRL · VLind | accuracy: 47.3 -> 55.3 | Improved |
| add consistency verification to data pipeline (ablation) | ViRL · V* | rationale accuracy: 55.3 -> 87.6 | Improved |
| add zoom-in dependency filtering (ablation) | ViRL · MMStar (4 core categories) | accuracy: 66.8 -> 67.1 | Improved |
| apply ViRL to Qwen2.5-VL-7B | Qwen2.5-VL-7B · VLind | accuracy: +8.7 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Deepeyes — DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026) · better on 3 of 3 shared results
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
- When Thinking Drifts: Evidential Grounding for Robust Video Reasoning (NeurIPS 2025 · similarity 0.90)
- DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Grounded Reinforcement Learning for Visual Reasoning (NeurIPS 2025 · similarity 0.89)
- See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs (CVPR 2026 · similarity 0.88)
- VES-RFT: Rewarding Visual Evidence Sensitivity to Mitigate Hallucinations in Large Vision-Language Models (CVPR 2026 · similarity 0.88)
- BLINK-Twice: You see, but do you observe? A Reasoning Benchmark on Visual Perception (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.
