AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Seeing to Ground: Visual Attention for Hallucination-Resilient MDLLMs

Vishal Narnaware, Animesh Gupta, Kevin Zhai, Zhenyi Wang, Shah Mubarak

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Decoding-time hallucination mitigation · 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 VISAGE entropy re-rankingMMaDA · MMMU-valaccuracy (%): 27.11 -> 29.44 (+8.59%)Improved
add VISAGE entropy re-rankingMMaDA · HallusionBenchaccuracy (%): 34.18 -> 36.83 (+7.75%)Improved
add VISAGE entropy re-rankingMMaDA · POPEF1: 75.97 -> 76.17 (+0.26%)No clear effect
add VISAGE entropy re-rankingMMaDA · MMEscore: 1383.29 -> 1372.05 (-0.81%)Hurt
add VCD (adapted to masked diffusion)MMaDA · MMMU-valaccuracy (%): 27.11 -> 28.44Improved
add VCD (adapted to masked diffusion)MMaDA · HallusionBenchaccuracy (%): 34.18 -> 34.80Improved
add VCD (adapted to masked diffusion)MMaDA · MMEscore: 1383.29 -> 1342.21Hurt
add VCD (adapted to masked diffusion)MMaDA · POPEF1: 75.97 -> 75.85No clear effect
set penalty alpha 0.3 rather than 0.5 (ablation)VISAGE · MMEscore: 1320.77 -> 1372.05Improved
set penalty alpha 0.3 rather than 0.1 (ablation)VISAGE · MMEscore: 1362.68 -> 1372.05Improved
use beta-quantile head aggregation instead of min (ablation)VISAGE · MMMU-valaccuracy (%): 28.56 -> 29.44Improved
use beta-quantile head aggregation instead of mean (ablation)VISAGE · MMMU-valaccuracy (%): 28.78 -> 29.44Improved

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