Table 5 Comparison of recognition accuracy with state-of-the-art methods on Something-Something v1 dataset
From: Fused behavior recognition model based on attention mechanism
Methods | Input modality | Pre_training | Top-1 val (%) | TOP-1test (%) |
|---|---|---|---|---|
TSN by ref. [23] (7 frames) | RGB | ImageNet | 18.48 | – |
MultiScale TRN [23] | RGB | ImageNet | 34.44 | 33.6 |
ECO (16 frames) [12] | RGB | ImageNet | 41.4 | – |
TRN (ResNet-50) by ref. [13] (8frames) | RGB | ImageNet | 38.9 | – |
ResNet34-3DRes18 (16 frames) | RGB | Kinetics | 41.012 | – |
Res34-SE-IM-Net (16 frames) | RGB | Kinetics | 41.398 | 36.5 |