AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Scaling Inference-Time Search with Vision Value Model for Improved Visual Comprehension

Xiyao Wang, Zhengyuan Yang, Linjie Li, Hongjin Lu, Yuancheng Xu, Chung-Ching Lin, Kevin Lin, Furong Huang, Lijuan Wang

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Bias-aware multi-image robustness · 23 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
guide search with VisVM instead of CLIP scoreLLaVA-Next-7B search · COCO Val2014 CHAIRsCHAIRs: 28.4 -> 26.2Improved
guide search with VisVM instead of greedy decodingLLaVA-Next-7B · COCO Val2014 CHAIRsCHAIRs: 32.4 -> 26.2Improved
guide search with VisVM instead of greedy decodingLLaVA-Next-7B · MMHalMMHal score: 2.94 -> 3.30Improved
use VisVM instead of MCTS searchLLaVA-Next-7B · COCO Val2014 CHAIRsCHAIRs: 25.9 vs 26.2No clear effect
train VisVM with SigLIP PRM instead of CLIP PRM (ablation)VisVM · COCO Val2014 CHAIRsCHAIRs: 26.2 -> 25.6Improved
select candidates with VisVM instead of CLIP from same candidates (ablation)LLaVA-Next-7B · COCO Val2014 CHAIRsCHAIRs: 31.6 -> 30.9Improved
increase search step size for VisVM-guided search (ablation)VisVM · COCO Val2014 CHAIRsCHAIRsImproved
apply VisVM to Qwen2-VL-7BQwen2-VL-7B · COCO Val2014 CHAIRsCHAIRs: 30.8 -> 24.5Improved
self-train on VisVM-search captions instead of greedy captionsLLaVA-Next-Mistral-7B · multimodal benchmark averageaverage improvement: -1.6% -> +10.8%Improved
self-train on greedy-decoding captionsLLaVA-Next-Mistral-7B · multimodal benchmark averageaverage improvement: -1.6%Hurt
self-train on VisVM captions instead of CLIP-PRM captionsQwen2-VL-7B-Instruct · multimodal benchmark averageaverage improvement: +5.0% -> +7.3%Improved
self-train on VisVM captionsLLaVA-Next-Mistral-7B · COCO Val2014 CHAIRsCHAIRs: 32.4 -> 22.6Improved

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