AI Paper Atlas

Research map / Vision-language and multimodal / Visual question answering

AVAM: a Universal Training-free Adaptive Visual Anchoring Embedded into Multimodal Large Language Model for Multi-image Question Answering

Kang Zeng, Guojin Zhong, Jintao Cheng, Jin Yuan, Zhiyong Li

ICCV 2025 · Vision-language and multimodal › Visual question answering · approach: Attention and prior-robust VQA · 4 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 Implicit Retrieval token pruning (R=0.1) (ablation)LLaVA-v1.5-7B · MuirBenchaccuracy (avg): 24.2 -> 26.0Improved
raise IR retention ratio R from 0.1 to 0.5 (ablation)LLaVA-v1.5-7B · MuirBenchaccuracy (avg): 26.0 -> 27.0Improved
add contrastive denoising-style CD (lambda=5) on top of IR (ablation)LLaVA-v1.5-7B · MuirBenchaccuracy (avg): 27.0 -> 27.6Improved
set CD lambda to 1 instead of 5 (ablation)LLaVA-v1.5-7B · MuirBenchaccuracy (avg): 27.6 -> 25.2Hurt
plug AVAM into LLaVA-v1.5-7BLLaVA-v1.5-7B · MuirBenchaccuracy (avg): 24.2 -> 27.6Improved
plug AVAM into InternVL2-8BInternVL2-8B · MuirBenchaccuracy (avg): 36.6 -> 39.5Improved
plug AVAM into Mantis-8BMantis-8B · MuirBenchaccuracy (avg): 34.0 -> 34.5Improved
plug AVAM into Qwen-VL-7BQwen-VL-7B · MuirBenchaccuracy (avg): 13.6 -> 14.0Improved
plug AVAM into LLaVA-v1.5-7BLLaVA-v1.5-7B · MIBench MIIaccuracy (avg): 16.2 -> 31.7Improved
plug AVAM into DeepSeek-VL-7BDeepSeek-VL-7B · MIBench MIIaccuracy (avg): 33.4 -> 52.0Improved
plug AVAM into LLaVA-OV-Qwen2-7BLLaVA-OV-Qwen2-7B · MIBench MIIaccuracy (avg): 67.2 -> 68.2Improved
plug AVAM into InternVL2-8BInternVL2-8B · Mantis-Evalaccuracy: 55.4 -> 59.7Improved

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