Benchmark RadarVIABench
VIABench evaluates multimodal large language models on three tasks from first-person videos of visually impaired individuals: proactive reminder, visual question answering, and vision-guided interaction. It includes 761 videos, 46.9 hours, and 14,526 annotations, with protocols for online and offline settings.
- Released
- 2026-07-16
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
General MLLMs are rarely tested for real-world assistance of blind users. VIABench focuses on tasks like anticipating navigation-critical events, which are underrepresented in existing benchmarks, providing a practical measure of model utility in assistive contexts.
Motivation
Visually impaired individuals (VIIs) encounter significant daily challenges due to limited access to visual information.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.