| ============================= A few notes on using the dataset ================================ | |
| Chen Chen, UNC-Charlotte | |
| https://webpages.uncc.edu/cchen62/ | |
| chen.chen@uncc.edu | |
| Cite our paper: | |
| Waqas Sultani, Chen Chen, Mubarak Shah, "Real-world Anomaly Detection in Surveillance Videos" | |
| IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2018 | |
| =============================================================================================== | |
| 1. Anomaly Detection Experiment | |
| - A) Videos: | |
| Anomaly-Videos-Part-1 -- Part-4 (4 zip files) | |
| Training-Normal-Videos-Part-1.zip and Training-Normal-Videos-Part-2.zip (normal videos for training) | |
| Testing_Normal_Videos.zip (normal videos for testing) | |
| UCF_Crimes-Train-Test-Split.zip contains the traing and testing split in our experiments | |
| (folder: Anomaly_Detection_splits) | |
| - B) Temporal Annotations for Testing Videos (anomaly videos) | |
| Temporal_Anomaly_Annotation_for_Testing_Videos.txt | |
| Each row of 'Temporal_Anomaly_Annotation.txt' is the annotation for a video, for example: | |
| Abuse028_x264.mp4 Abuse 165 240 -1 -1 | |
| - The first column is the name of the video | |
| - The second column is the name of the anomalous event | |
| - The third column is the starting frame of the event (you will have to convert each video to image frames first) | |
| - The fourth column is the ending frame of the event. | |
| - For videos in which second instance of event occurs, fifth and sixth contains starting and ending frames of second instance. | |
| Negative number means no anomalous event instance. In this example, abuse (instance) only occurs once. | |
| Note: Ours videos have 30 frames per second. | |
| 2. Anomaly Event Recognition Experiment | |
| Classify 13 anomaly events and normal event | |
| Normal_Videos_for_Event_Recognition.zip contains the normal videos we used for event recognition experiment | |
| Note: rename the unziped folder to "Normal_Videos_event" | |
| UCF_Crimes-Train-Test-Split.zip also contains the traing and testing split for event recognition experiment | |
| (folder: Action_Regnition_splits) | |