Table 2 Comparison with existing works on the UCF-Crime dataset.
From: Weakly supervised video anomaly detection based on hyperbolic space
Supervision | Method | Feature | T | AUC (%) | Parameters (M) |
|---|---|---|---|---|---|
Un- supervision | GODS (2019)34 | \(I3D^{RGB}\) | – | 70.46 | – |
GCL(2022)1 | ResNext | – | 71.04 | \(3.44^{*}\) | |
DyAnNet (2023)27 | \(I3D^{RGB}\) | 32 | 79.76 | \(1.09^{*}\) | |
C2FPL (2024)35 | \(I3D^{RGB}\) | – | 80.65 | 2.13 | |
CLAP (2024)36 | \(I3D^{RGB}\) | – | 80.90 | 2.13 | |
Weakly- supervision | Sultani et al.(2018)2 | \(C3D^{RGB}\) | 32 | 77.92 | 2.11 |
GCL (2022)1 | ResNext | – | 79.84 | \(3.44^{*}\) | |
GCN (2021)17 | \(C3D^{RGB}\) | 32 | 81.08 | 2.17 | |
GCN (2021)17 | \(TSN^{RGB}\) | 32 | 82.12 | 2.17 | |
GCN (2021)17 | \(TSN^{FLOW}\) | 32 | 78.08 | 2.17 | |
MIST (2021)6 | \(I3D^{RGB}\) | 32 | 82.30 | 30.99 | |
HL-NET(2020)18 | \(I3D^{RGB}\) | 200 | 82.44 | 0.84 | |
CLAWS (2021)37 | \(C3D^{RGB}\) | – | 83.03 | – | |
RTFM (2021)4 | \(I3D^{RGB}\) | 32 | 84.30 | 24.72 | |
Cao et al. (2022)38 | \(I3D^{RGB}\) | 150 | 84.67 | 2.17 | |
S3R (2022)32 | \(I3D^{RGB}\) | 32 | 85.99 | 81.44 | |
MGFN (2022)5 | \(I3D^{RGB}\) | 32 | 86.98 | 28.65 | |
UR-DMU (2023)7 | \(I3D^{RGB}\) | 200 | 86.97 | 6.49 | |
Ours | \(I3D^{RGB}\) | 200 | 85.21 | 0.61 |
