Tag
Introduces VIABench, a comprehensive video benchmark for evaluating multimodal large language models in real-world visual assistance for blind and visually impaired individuals, covering 761 videos and 14,526 annotations across three tasks.