PaSBench-Video is a curated video benchmark for proactive safety and risk anticipation. Each sample pairs a video with frame/time annotations and reviewed risk-response prompts.
metadata.csv: Hugging Face VideoFolder metadata. The file_name column points to the video file.
videos/: 481 curated video files grouped by original source dataset.
annotations/final_annotations.json: sanitized public annotations used to generate… See the full description on the dataset page:
https://huggingface.co/datasets/beingbetter11643/PaSBench-Video.
