Benchmark RadarPaSBench-Video
PaSBench-Video is a 740-video benchmark for proactive safety warning with 481 risk and 259 no-risk videos across driving, healthcare, daily life, and industrial production. Annotations include frame-level risk onset and accident boundaries. Models must process video causally and output a warning that is temporally calibrated and content-correct.
- Released
- 2026-06-01
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
This benchmark addresses a gap in evaluating video MLLMs for real-time safety monitoring, emphasizing temporal calibration and false-positive control on safe scenes. It provides a standardized protocol to compare models' ability to issue timely warnings, which is critical for deployment in safety-sensitive applications.
Motivation
Between the first visible sign of danger and the moment an accident occurs, there is often a window where intervention remains possible.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.