AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

VisReason: A Large-Scale Dataset for Visual Chain-of-Thought Reasoning

Lingxiao Li, Yifan Wang, Xinyan Gao, Chen Tang, Xiangyu Yue, Chenyu You

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
fine-tune on VisReason-Pro with adaptive zoom-inQwen2.5-VL-7B · Visual-CoT averageaccuracy: 0.770 -> 0.802Improved
fine-tune on VisReason instead of no fine-tuning (ablation)Qwen2.5-VL-7B · Visual-CoT relation reasoningaccuracy: 0.598 -> 0.678Improved
fine-tune on VisReason instead of no fine-tuning (ablation)Qwen2.5-VL-7B · Visual-CoT Doc/Textaccuracy: 0.920 -> 0.864Hurt
add adaptive zoom-in strategy (ablation)VisReason-7B · Visual-CoT averageaccuracy: 0.777 -> 0.789Improved
add adaptive zoom-in strategy (ablation)VisReason-7B + Pro · Visual-CoT averageaccuracy: 0.780 -> 0.802Improved
add VisReason-Pro data to VisReason (ablation)VisReason-7B · Visual-CoT fine-grainedaccuracy: 0.798 -> 0.809Improved
add VisReason-Pro data to VisReason (ablation)VisReason-7B · Visual-CoT averageaccuracy: 0.777 -> 0.780No clear effect
add VisReason-Pro data to VisReason (ablation)VisReason-7B with AZ · Visual-CoT Doc/Textaccuracy: 0.882 -> 0.892Improved
evaluate with direct QA instead of multi-round CoTVisReason-Pro-7B · Visual-CoT Text/Docaccuracy: 0.892 -> 0.916Improved
evaluate with direct QA instead of multi-round CoTVisReason-Pro-7B · Visual-CoT spatial relationaccuracy: 0.722 -> 0.599Hurt
evaluate with direct QA instead of multi-round CoTVisReason-Pro-7B · V*Benchaccuracy: 0.603 -> 0.791Improved
evaluate with direct QA instead of multi-round CoTVisReason-Pro-7B · MMEscore: 0.777 -> 0.856Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy