OmniFall: A Unified Benchmark for Staged-to-Wild Fall Detection
OmniFall is a comprehensive fall detection benchmark with dense temporal segment annotations across three components: OF-Staged (8 public lab datasets), OF-In-the-Wild (genuine accidents from OOPS), and OF-Synthetic (12,000 diffusion-generated videos with demographic diversity). All components share a sixteen-class activity taxonomy.
[Paper] [Project Page]
Quickstart
Labels… See the full description on the dataset page: https://huggingface.co/datasets/xxb666/omnifall.
