Research map / Vision-language and multimodal / Multimodal LLMs
Awakening Visual Reasoning: Mitigating Post-Training Failure in Vision-Text Compression
Xing Xi, Liyao Li, Hao Chen, Ningtao Wang, Peixian Chen, Peilin Tong, Xing Fu, Yu Cheng, Haobo Wang, Gang Chen, Junbo Zhao, Ronghua Luo
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Studies why vision-text compression (rendering text as dense visual tokens) degrades reasoning in VLMs even though OCR fidelity stays high, and proposes a mitigation for this post-training failure.
Problem
Vision-text compression renders long text into dense visual tokens for scalable long-context multimodal modeling, but VLMs that reason well on native text fail on visually compressed equivalents, especially in long-range retrieval and multi-step reasoning.
Approach
Identifies a reasoning gap between native-text and visually compressed inputs despite high OCR fidelity, and targets post-training as the source of the failure; the abstract does not specify the method in detail.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CoRe alignment to RL on visual prompts | Visual GRPO · Math average (visual) | accuracy (%): 29.44 -> 42.97 | Improved |
| add CoRe alignment to SFT on visual prompts | Visual SFT · Math average (visual) | accuracy (%): 15.42 -> 20.92 | Improved |
| add CoRe alignment to RL | Visual GRPO · GSM8K (visual) | accuracy (%): 67.31 -> 82.19 | Improved |
| add CoRe alignment to RL | Visual GRPO · HotpotQA (visual) | accuracy (%): 9.00 -> 23.00 | Improved |
| add CoRe alignment to RL | Visual GRPO · 2WikiMultihopQA (visual) | accuracy (%): 13.00 -> 35.00 | Improved |
| add CoRe alignment to RL | Visual GRPO · InfoTabs (visual) | accuracy (%): 61.29 -> 70.18 | Improved |
| add CoRe alignment to RL | Visual GRPO · HiTab (visual) | accuracy (%): 3.16 -> 5.06 | Improved |
| add CoRe alignment to RL | Visual GRPO · Qasper (visual) | accuracy (%): 1.00 -> 4.00 | Improved |
| apply standard GRPO to visual prompts instead of text | Visual GRPO · HotpotQA | accuracy (%): 7.0% to 9.0% | Improved |
| use ground-truth CoT alignment (CoRe-SFT) rather than one self-generated rollout (K = 1) (ablation) | CoRe · AIME 2024 | accuracy (%): 3.33 -> 6.67 | Improved |
| use self-generated rollout (K = 1) rather than ground-truth trace (ablation) | CoRe · MATH-500 | accuracy (%): 42.20 -> 42.80 | No clear effect |
| increase rollouts from K = 1 to K = 4 (ablation) | CoRe-RL · MATH-500 | accuracy (%): 42.80 -> 47.20 | Improved |
Surprising findings
- VLMs with high OCR decoding fidelity on compressed visual text still show a severe reasoning gap relative to native text.
Closest papers by meaning
- VisionPulse: Dynamic Visual Sparsity for Efficient Multimodal Reasoning (ICML 2026 · similarity 0.90)
- Activating Visual Context and Commonsense Reasoning Through Masked Prediction in VLMs (AAAI 2026 · similarity 0.90)
- More Thought, Less Accuracy? On the Dual Nature of Reasoning in Vision-Language Models (ICLR 2026 · similarity 0.90)
- SPARC: Separating Perception And Reasoning Circuits for Test-time Scaling of VLMs (ICML 2026 · similarity 0.90)
- SLVR: Structured Latent Visual Reasoning via Human-like Reasoning Flows (NeurIPS 2026 · similarity 0.90)
- Reinforcing Multimodal Reasoning Against Visual Degradation (NeurIPS 2026 · similarity 0.90)
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.
