AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Staying VIGILant: Mitigating Visual Laziness in MLLMs via Information-Theoretic Alignment

Xi Xiao, Chen Liu, Chih-Ting Liao, Yunbei Zhang, Qizhen Lan, YUXIANG WEI, Lin Zhao, Janet Wang, Jianyang Gu, Muchao Ye, Tianyang Wang, Hao Xu

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Reinforcement learning for visual reasoning

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 visual anchor (counterfactual blind-state constraint) (ablation)VIGIL on Qwen2.5-VL-7B · POPE RandomF1: 85.3 -> 88.5Improved
add visual anchor (ablation)VIGIL on Qwen2.5-VL-7B · AMBER GenerativeAMBER score: 43.1 -> 46.5Improved
add hard negatives (ablation)VIGIL on Qwen2.5-VL-7B · POPE RandomF1: 86.5 -> 88.5Improved
add dynamic gating (ablation)VIGIL on Qwen2.5-VL-7B · POPE RandomF1: 87.4 -> 88.5Improved
add dynamic gating (ablation)VIGIL on Qwen2.5-VL-7B · MathVistaaccuracy: 49.2 -> 49.5Improved
add visual anchor (ablation)VIGIL on Qwen2.5-VL-7B · MathVistaaccuracy: 49.0 -> 49.5Improved
replace difficulty reweighting (DA-DPO) with explicit attention-masking CVD (ablation)DPO on Qwen2.5-VL-7B · POPE RandomF1: 85.7 -> 88.5Improved
filter easy samples (10%) instead of masking (ablation)DPO on Qwen2.5-VL-7B · POPE RandomF1: 82.8 -> 84.3Improved
VIGIL vs DPO on high visual-dependency samplesQwen2.5-VL-7B · POPE Random, High VDI bucketF1: 78.2 -> 87.5Improved
VIGIL vs DPO on low visual-dependency samplesQwen2.5-VL-7B · POPE Random, Low VDI bucketF1: 88.5 -> 89.2No clear effect
scale base model from 7B to 72BVIGIL · POPE Adversarialgain over SFT (points): 4.1 -> 5.3Improved
apply VIGIL to localization (zero-shot referring expression)Qwen2.5-VL-7B · RefCOCOg valAcc@0.5: 44.8 -> 49.5Improved

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