Anomaly Detection on UCF-Crime normal (test)
0.5False Alarm RateTemporal encoding-decoding network
Evaluation results
| Method | Links | |
|---|---|---|
| Temporal encoding-decoding networkFeatures=C3D, Threshold=50%2020.07 | 0.5 | |
| Proposed Method (3D ResNet+constr.+new rank.loss)Backbone=3D ResNet, Constraints=Yes, Loss Function=New ranking loss, Threshold=50%2020.02 | 0.8 | |
| Proposed Method (3D ResNet+constr.+loss)Backbone=3D ResNet, Constraints=Yes, Loss Function=Original ranking loss, Threshold=50%2020.02 | 0.83 | |
| Deep MIL RankingThreshold=50%2018.01 | 1.9 | |
| W. Sultani et al.Threshold=50%2020.02 | 1.9 | |
| Sultani et al.Features=C3D, Threshold=50%2020.07 | 1.9 | |
| Zaheer et al.Features=C3D, Threshold=50%2020.07 | 2.1 | |
| Zhong et al.Features=C3D, Threshold=50%2020.07 | 2.8 | |
| Lu et al.Threshold=50%2018.01 | 3.1 | |
| C. Lu et al.Threshold=50%2020.02 | 3.1 | |
| Hassan et al.Features=C3D, Threshold=50%2020.07 | 3.1 | |
| Hasan et al.Threshold=50%2018.01 | 27.2 | |
| M. Hasan et al.Threshold=50%2020.02 | 27.2 | |
| Li et al.Features=C3D, Threshold=50%2020.07 | 27.2 |