Research Article
Anomaly Detection in Videos Using Two-Stream Autoencoder with Post Hoc Interpretability
Table 3
Comparison of area under ROC curve (frame-level AUC) of different methods.
| Method | Avenue | Ped2 | Subway Entrance | Subway Exit |
| ST-AE [20] | 76.5 | 81.7 | 81.8 | 86.4 | Unmasking [10] | 80.6 | 82.2 | 70.6 | 85.7 | GMFC-VAE [11] | 78.6 | 84.9 | 83.7 | 87.4 | RBM [12] | 78.7 | 86.4 | — | — | DAEs + cGAN [14] | 73.6 | 86.1 | 84.1 | 87.3 | ConvAe [7] | 74.3 | 79.7 | 84.9 | 83.9 | Conv-LSTM-AE [18] | 76.4 | 82.9 | 83.3 | 86.4 | Spatial model | 78.1 | 83.8 | 84.7 | 90.2 | Temporal model | 77.8 | 84.0 | 85.0 | 85.4 | Spatial + temporal | 80.3 | 84.5 | 87.3 | 90.8 |
|
|
Higher AUC is better.
|