SPARC (Selective Progressive Attention ReCalibration) addresses a critical limitation in MLLMs: weakening visual attention during lengthy caption generation.

• Maintains consistent visual attention throughout the entire caption generation process
• Delivers more precise and detailed image descriptions without requiring additional training
• Achieves better balance between precision and recall in image captioning
• Works as a training-free solution that can enhance existing models

Medical Impact: By generating more detailed and accurate image descriptions, SPARC significantly improves assistive technologies for visually impaired individuals, enabling better understanding of visual content in medical and everyday contexts.

Visual Attention Never Fades: Selective Progressive Attention ReCalibration for Detailed Image Captioning in Multimodal Large Language Models