Table 6 Comparison of the complexity and accuracy between our methods and state-of-the-art methods on the HMDB51 and UCF101 datasets
From: Fused behavior recognition model based on attention mechanism
Methods | FLOPs | Param | Depth | VPS | HMDB51 (%) | UCF101 (%) |
|---|---|---|---|---|---|---|
I3D(RGB) [9] | 139.39G | 12.7 M | 72 | 0.5 | 74.5 | 95.4 |
ResNext-101 [19] | 192.31G | 60.63 M | 101 | – | 63.8 | 90.7 |
ResNet-101 [19] | 277.23G | 86.92 M | 101 | – | 61.7 | 88.9 |
ResNet34-3DRes18 (16 frames) | 85.57G | 55.78 M | 48 | 20.2 | 70.997 | 92.143 |
Res34-SE-IM-Net (16 frames) | 85.6G | 60.2 M | 48 | 18.8 | 71.85 | 92.196 |