Hart

New HART technique enables LMMs to reason with high-resolution images without annotations

Researchers have developed a new technique called HART (High-resolution Annotation-free Reasoning Technique) to improve how Large Multimodal Models (LMMs) handle high-resolution images. Current LMMs struggle with the la…