Table 1 AP comparison on the XD-Violence dataset.
From: Weakly supervised video anomaly detection based on hyperbolic space
Supervision | Method | Feature | T | AP (%) | Parameters (M) |
|---|---|---|---|---|---|
Un- supervision | SVM baseline | – | – | 50.78 | – |
Hasan et al.28 | – | – | 30.77 | – | |
OCSVM29 | – | – | 27.25 | – | |
Weakly-supervision | Sultani et al. (2018)2 | \(C3D^{RGB}\) | 32 | 73.20 | 2.11 |
HL-NET (2020)18 | \(I3D^{RGB}\) | 200 | 73.67 | 0.84 | |
HL-NET (2020)18 | \(I3D^{RGB} \& VGGish\) | 200 | 78.64 | 0.84 | |
RTFM (2021)4 | \(I3D^{RGB}\) | 32 | 77.81 | 24.72 | |
MSL (2022)30 | \(VideoSwin^{RGB}\) | 32 | 78.28 | – | |
CU-NET(2023)31 | \(I3D^{RGB}\) | 32 | 78.74 | 2.11 | |
CU-NET(2023)31 | \(I3D^{RGB} \& VGGish\) | 32 | 81.43 | 2.11 | |
MGFN (2022)5 | \(I3D^{RGB}\) | 32 | 79.19 | 28.65 | |
S3R (2022)32 | \(I3D^{RGB}\) | 32 | 80.26 | 81.44 | |
UR-DMU (2023)7 | \(I3D^{RGB}\) | 200 | 81.66 | 6.49 | |
UR-DMU (2023)7 | \(I3D^{RGB} \& VGGish\) | 200 | 81.77 | 6.49 | |
Pang et al.(2021)33 | \(I3D^{RGB} \& VGGish\) | – | 81.69 | – | |
Ours | \(I3D^{RGB}\) | 150 | 82.67 | 0.61 |
