DVD: A Comprehensive Dataset for Advancing Violence Detection in Real-World Scenarios

TL;DR

This paper introduces DVD, a large-scale, richly annotated video dataset designed to improve violence detection models by providing diverse, detailed, and realistic data for training and evaluation.

Contribution

The creation of DVD, a comprehensive dataset with extensive annotations, diverse scenarios, and rich metadata to advance violence detection research.

Findings

DVD enables better model generalization

Improves violence detection accuracy

Supports diverse real-world scenarios

Abstract

Violence Detection (VD) has become an increasingly vital area of research. Existing automated VD efforts are hindered by the limited availability of diverse, well-annotated databases. Existing databases suffer from coarse video-level annotations, limited scale and diversity, and lack of metadata, restricting the generalization of models. To address these challenges, we introduce DVD, a large-scale (500 videos, 2.7M frames), frame-level annotated VD database with diverse environments, varying lighting conditions, multiple camera sources, complex social interactions, and rich metadata. DVD is designed to capture the complexities of real-world violent events.