Research map / Vision-language and multimodal / Visual question answering
Deeper Thought, Weaker Aim: Understanding and Mitigating Perceptual Impairment during Reasoning in Multimodal Large Language Models
Ruiying Peng, Xueyu Wu, Jing Lei, Lu Hou, Yuanzheng Ma, Xiao-Hui Li
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 VRGA attention reweighting | Qwen2.5-VL-3B · HaloQuest | comprehensive score S: 0.445 -> 0.488 | Improved |
| add VRGA attention reweighting | Qwen2.5-VL-3B · HaloQuest | irrelevance degree I: 0.601 -> 0.405 | Improved |
| add VRGA attention reweighting | Qwen2.5-VL-7B · HaloQuest | accuracy (%): 66.67 -> 73.96 | Improved |
| add VRGA attention reweighting | Qwen2.5-VL-7B · HaloQuest | comprehensive score S: 0.502 -> 0.549 | Improved |
| add VRGA attention reweighting | Qwen3-VL-30B · HaloQuest | accuracy (%): 69.20 -> 68.05 | Hurt |
| add VRGA attention reweighting | Qwen3-VL-30B · MMStar | accuracy (%): 66.1 -> 67.1 | Improved |
| add VRGA attention reweighting | Qwen2-VL-7B · HallusionBench | accuracy (%): 55.10 -> 57.10 | Improved |
| add VRGA attention reweighting | Qwen2-VL-7B · MMStar | accuracy (%): 52.37 -> 51.07 | Hurt |
| add CCoT reasoning method | Qwen2-VL-7B · HaloQuest | accuracy (%): 53.81 -> 29.30 | Hurt |
| add ICoT reasoning method | Qwen2-VL-7B · HaloQuest | accuracy (%): 53.81 -> 47.85 | Hurt |
| mask heads chosen by entropy-focus criterion instead of no masking (ablation) | Qwen2.5-VL-3B · TextVQA | accuracy (%): 87.64 -> 24.31 | Hurt |
| mask random heads instead of no masking (ablation) | Qwen2.5-VL-3B · TextVQA | accuracy (%): 87.64 -> 83.38 | Hurt |
Closest papers by meaning
- Look-Back: Implicit Visual Re-focusing in MLLM Reasoning (AAAI 2026 · similarity 0.92)
- From Narrow to Panoramic Vision: Attention-Guided Cold-Start Reshapes Multimodal Reasoning (ICLR 2026 · similarity 0.91)
- Thinking Diffusion: Penalize and Guide Visual-Grounded Reasoning in Diffusion Multimodal Language Models (CVPR 2026 · similarity 0.91)
- Where Does Vision Meet Language? Understanding and Refining Visual Fusion in MLLMs via Contrastive Attention (CVPR 2026 · similarity 0.91)
- Seeing but Not Believing: Probing the Disconnect Between Visual Attention and Answer Correctness in VLMs (ICLR 2026 · similarity 0.91)
- LENS: Multi-level Evaluation of Multimodal Reasoning with Large Language Models (ICLR 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.
