CHAD

该数据集名为CHAD，是一个高分辨率多摄像头数据集，包含了大约115万帧画面，其中有109万帧正常画面和5.9万帧异常画面，并对人的检测、追踪和姿态进行了详细标注。CHAD模拟了一个真实的停车场监控环境，使用四个高分辨率摄像头以每秒30帧的速度进行录制，并包含了多种异常类别。该数据集的规模大约为115万帧，主要针对视频异常检测（VAD）任务。

The dataset named CHAD is a high-resolution multi-camera dataset containing approximately 1.15 million frames, of which 1.09 million are normal frames and 59,000 are abnormal frames, with detailed annotations for human detection, tracking and pose estimation. CHAD simulates a real-world parking lot surveillance environment, where four high-resolution cameras capture footage at 30 frames per second and cover multiple anomaly categories. With a total scale of around 1.15 million frames, this dataset is primarily designed for video anomaly detection (VAD) tasks.

搜集汇总

背景与挑战

背景概述

CHAD是一个高分辨率多摄像头监控视频异常检测数据集，提供412个视频约1.15M帧，包含边界框、Re-ID、姿态注释和帧级异常标签，适用于有监督和无监督学习。