AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Integrating Visual Interpretation and Linguistic Reasoning for Geometric Problem Solving

Zixian Guo, Ming Liu, Qilong Wang, Zhilong Ji, Jinfeng Bai, Lei Zhang, Wangmeng Zuo

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 1 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
add GPT-4o description preprocessing (ablation)Ours SFT · MathVerse Text Dominantaccuracy: 47.9 -> 52.3Improved
add GPT-4o description preprocessing (ablation)Ours SFT · MathVerse Vision Onlyaccuracy: 36.7 -> 39.7Improved
add GPT-4o description preprocessing (ablation)Ours SFT · MathVerse Vision Dominantaccuracy: 38.8 -> 42.6Improved
use decoupled interpreter and reasoner instead of single LVLM (ablation)Qwen2VL-7B · MathVerse Vision Dominantaccuracy: 39.1 -> 47.2Improved
use decoupled interpreter and reasoner instead of single LVLM (ablation)Qwen2VL-7B · MathVerse Vision Onlyaccuracy: 37.3 -> 43.8Improved
use decoupled interpreter and reasoner instead of single LVLM (ablation)Qwen2VL-7B · MathVerse Text Dominantaccuracy: 51.1 -> 54.3Improved
add stage 2 RL of LVLM (ablation)Ours after SFT · MathVerse Vision Onlyaccuracy: 39.7 -> 42.8Improved
add stage 2 RL of LVLM (ablation)Ours after SFT · MathVerse Text Dominantaccuracy: 52.3 -> 51.5Hurt
add stage 3 RL of LLM (ablation)Ours after SFT · MathVerse Text Liteaccuracy: 46.9 -> 49.8Improved
add stage 3 RL of LLM on top of stage 2 (ablation)Ours stage 1+2 · MathVerse Vision Dominantaccuracy: 45.5 -> 47.2Improved
add stage 3 RL of LLM on top of stage 2 (ablation)Ours stage 1+2 · MathVerse Text Liteaccuracy: 49.8 -> 49.0Hurt
decoupled reasoning vs LLaVA-CoT end-to-endLLaVA-CoT-11B · MathVerseaccuracy: 36.2 -> 42.5Improved

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.

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

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