AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Multiple Images Distract Large Multimodal Models via Attention Fragmentation

Tingrui Qiao, Di Zhao, Yuzhuo Li, Bo Pang, Caroline Walker, Chris Cunningham, Yun Sing Koh

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Training-free token reduction for MLLMs · 0 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 Attention Remasking (AR)Qwen3-VL-8B · MMIUaccuracy (%): 55.6 -> 59.4Improved
add Attention Remasking (AR)Qwen3-VL-8B · MuirBenchaccuracy (%): 51.3 -> 54.9Improved
add Attention Remasking (AR)LLaVA-OneVision-7B · MMIUaccuracy (%): 36.9 -> 41.2Improved
add Attention Remasking (AR)InternVL3.5-8B · MMIUaccuracy (%): 49.3 -> 54.1Improved
add Attention Remasking (AR)DeepSeek-VL2-Small · MIRBaccuracy (%): 45.8 -> 50.7Improved
add Attention Remasking (AR)Qwen3-VL-32B · MMIUaccuracy (%): 59.8 -> 62.6Improved
add Attention Remasking (AR)DeepSeek-VL2-27B · MuirBenchaccuracy (%): 43.7 -> 46.8Improved
add VAR post-softmax redistributionQwen3-VL-8B · MMIUaccuracy (%): 55.6 -> 55.9No clear effect
add SoFA attention interpolationQwen3-VL-8B · MMIUaccuracy (%): 55.6 -> 56.7Improved
take grounding prior from last layer instead of layer 1 (ablation)AR on Qwen3-VL-8B · MMIUaccuracy (%): 59.4 -> 53.2Hurt
take grounding prior from last layer instead of layer 1 (ablation)AR on InternVL3.5-8B · MMIUaccuracy (%): 54.1 -> 46.5Hurt
take grounding prior from middle layer instead of layer 1 (ablation)AR on Qwen3-VL-8B · MMIUaccuracy (%): 59.4 -> 54.8Hurt

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