04:00
2026-06-12
arxiv.org
computer-vision
Magnifying What Matters: Attention-Guided Adaptive Rendering for Visual Text Comprehension
Researchers have identified that vision-language models (VLMs) often locate relevant text in images but fail to utilize it for answering questions, a phenomenon called "localization-without-utilizatio…