UCF-Crime
收藏资源简介:
UCF-Crime数据集是一个用于视频异常检测的大规模资源,包含了长时间、未剪辑的监控视频,其中包含时间局部化的异常事件。数据集广泛用于帧级别和视频级别的评估。UCF-Crime数据集由首尔大学人工智能与机器人系创建,旨在解决视频异常检测问题,特别是针对监控和公共安全领域。数据集包含6,130个帧级别的问答对和2,840个视频级别的问答对,共计8,970个样本,涵盖了不同类型的异常事件。
UCF-Crime Dataset is a large-scale resource for video anomaly detection, which contains untrimmed long-duration surveillance videos with temporally localized anomalous events. It is widely used for frame-level and video-level evaluations. Developed by the Department of Artificial Intelligence and Robotics at Seoul National University, this dataset was created to address video anomaly detection problems, especially in the fields of surveillance and public safety. The dataset includes 6,130 frame-level question-answer pairs and 2,840 video-level question-answer pairs, totaling 8,970 samples covering various types of anomalous events.
- 1DUAL-VAD: Dual Benchmarks and Anomaly-Focused Sampling for Video Anomaly Detection首尔大学人工智能与机器人系 · 2025年



