AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Taming the Untamed: Graph-Based Knowledge Retrieval and Reasoning for MLLMs to Conquer the Unknown

Bowen Wang, Zhouqiang Jiang, Yasuaki Susumu, Shotaro Miwa, Tianwei Chen, Yuta Nakashima

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning · 2 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 graph knowledge retrieval (multi-agent, unaided-online) over vanillaGPT-4o · MH Benchmarkaccuracy: .3122 -> .5105Improved
add human captions instead of images (Vanilla+)GPT-4o · MH Benchmarkaccuracy: .3122 -> .3924Improved
provide annotated subgraph (Knowledgeable)GPT-4o · MH Benchmarkaccuracy: .3924 -> .8565Improved
generate captions online during retrieval rather than offlineGPT-4o · MH Benchmarkaccuracy: .4050 -> .5105Improved
generate captions online rather than offlineGPT-4o mini · MH Benchmarkaccuracy: .4514 -> .3544Hurt
generate captions online rather than offlineClaude 3.7 Sonnet · MH Benchmarkaccuracy: .3628 -> .4388Improved
use human-selected keyframes instead of equal-interval sampling (ablation)GPT-4o · MH Benchmark Unaided-Onlineaccuracy: .4840 -> .5105Improved
add monster name to question (ablation)GPT-4o Unaided-Online · MH Benchmarkaccuracy: .3781 -> .5105Improved
add extra information to question (ablation)GPT-4o Unaided-Online · MH Benchmarkaccuracy: .4075 -> .5105Improved
use InternVL as Validation agent instead of GPT-4o (ablation)GPT-4o pipeline · MH Benchmarkaccuracy: .5105 -> .4128Hurt
use InternVL as Summarizer instead of GPT-4o (ablation)GPT-4o pipeline · MH Benchmarkaccuracy: .5105 -> .4864Hurt
replace offline captions with InternVideo2.5 captions (ablation)GPT-4o Unaided-Offline · MH Benchmarkaccuracy: .4050 -> .3697Hurt

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