About this document
ViDAS: Video Danger Assessment Dataset by namannanda0 is a document available to read on EtoBox.
The document presents ViDAS, a novel dataset for assessing danger in video content, featuring 100 YouTube videos annotated with human-assigned danger ratings from 0 to 10. It explores the effectiveness of Large Language Models (LLMs) in evaluating danger levels and compares their performance to human assessments using Mean Squared Error metrics. The research aims to establish a standardized benchmark for danger assessment, improve understanding of human and LLM danger perception, and identify new research d
- Author
- namannanda0
- Language
- EN