October 7, 2025Open Access

MLLMs Know Where to Look: Training-free Perception of Small Visual Details with Multimodal LLMs

View Full Paper

Ask AI

Bookmark

MLLMs struggle with small visual details, leading to performance sensitivity based on image subject size.
Their attention patterns indicate they often know where to look, even if their answers are incorrect.
Training-free intervention methods using attention and gradient maps can enhance MLLMs' visual perception.
Evaluations on multiple benchmarks show significant accuracy improvements without additional training.

Discussion

No takes yet. Share an insight, caveat, or question.

Intervention study reveals MLLMs struggle with small visual details, suggesting novel training-free methods to enhance perception.

Zhang et al. (2025) studied this question.

View Full Paper

Ask AI

Bookmark