LensVLM enhances Vision Language Models by selectively expanding context for compressed visual text. This approach improves accuracy when text is rendered at resolutions too low for standard encoders, enabling better understanding of visually represented text.
Opening Kapyn…