EgoBlind introduces a groundbreaking dataset of 1,210 egocentric videos recorded by blind individuals, designed to evaluate and improve multimodal large language models for visual assistance.

• First-of-its-kind collection featuring 4,927 questions posed by blind individuals
• Captures authentic daily activities from a first-person perspective
• Focuses on real-world assistance needs across various scenarios
• Provides a foundation for developing more helpful AI visual assistants

Medical Impact: This research addresses critical accessibility challenges in healthcare by enabling the development of AI systems that can provide real-time visual guidance for blind individuals, enhancing their autonomy in medical settings and daily life activities.

EgoBlind: Towards Egocentric Visual Assistance for the Blind People