AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

MMCR: Benchmarking Cross-Source Reasoning in Scientific Papers

Yang Tian, Zheng Lu, Mingqi Gao, Zheng Liu, Bo Zhao

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 3 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
use circular evaluation (5 option rotations) instead of single-pass (ablation)MMCR evaluation · MMCRaccuracy: MiniCPM-o 2.6 drops to approximately 36% of its non-circular score; random 20% -> 0.2^5 x 100%Hurt
raise rendering DPI from 36 to 108 (ablation)VLMs on MMCR · MMCRaccuracy: steady improvements as DPI increased to 108Improved
raise rendering DPI above 108 up to 180 (ablation)VLMs on MMCR · MMCRaccuracy: performance fluctuates and even declines when DPI exceeds 108Hurt
add CoT prompt to small VLMs (ablation)small VLMs (e.g. MiniCPM-o 2.6) · MMCRaccuracy: notable performance declineHurt
add CoT prompt to large VLMs (ablation)InternVL2.5-78B and Qwen2.5-VL-72B · MMCRaccuracy: performance improved with CoT promptImproved
add CoT prompt to Idefics3-8B (ablation)Idefics3-8B · MMCRaccuracy: slight improvement; model still answered directlyNo clear effect
proprietary GPT-4o versus best open-source Qwen2.5-VL-72BGPT-4o · MMCR overallaccuracy: 39.86 -> 48.55 (8.69% advantage)Improved
scale Qwen2.5-VL from 7B to 72BQwen2.5-VL · MMCR overallaccuracy: 21.74 -> 39.86Improved
scale InternVL2.5 from 8B to 78BInternVL2.5 · MMCR overallaccuracy: 12.32 -> 14.49Improved
table questions versus figure questionsevaluated VLMs · MMCRaccuracy: 72.22% (13/18) of models equal or worse on table than figure questionsHurt
concatenate multi-page images into one image for models without multi-image supportVLMs marked with * · MMCR overallaccuracy: concatenated models score 0.36-4.71 overallHurt

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