metadata
pipeline_tag: video-classification
PRE-ACT: Progressive Risk Estimation for Accident Anticipation
This repository hosts the model and code for the paper Progressive Risk Estimation for Accident Anticipation.
The model estimates a continuously evolving accident risk signal from driving videos, progressively increasing the risk as a crash approaches while suppressing premature false alarms. It is designed for video-based accident anticipation and achieves significant improvements on MM-AU subsets and Nexar.
For code, training details, and inference scripts, please refer to the GitHub repository.