Research Article
Integrating Temporal and Spatial Attention for Video Action Recognition
Table 1
Network structure of T-CNN.
| | Input: 3 × 16 × 224 × 224 |
| | Stage 1 | Conv 3–32 + BN + ReLU | | Conv 32–32 + BN + ReLU |
| | Stage 2 | Conv 32–64 + BN + ReLU | | Conv 64–64 + BN + ReLU |
| | Stage 3 | Conv 64–96 + BN + ReLU | | Conv 96–96 + BN + ReLU |
| | Stage 4 | Conv 96–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Dilation Conv(4) 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Dilation Conv(4) 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Dilation Conv(4) 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Conv 160–160 + BN + ReLU | | Dilation Conv(4) 160–160 + BN + ReLU |
| | Stage 5 | Conv 160–224 + BN + ReLU | | Conv 224–224 + BN + ReLU | | Conv 224–224 + BN + ReLU | | Dilation Conv(2) 224–224 + BN + ReLU | | Conv 224–224 + BN + ReLU | | Dilation Conv(4) 224–224 + BN + ReLU | | Conv 224–224 + BN + ReLU | | Dilation Conv(4) 224–224 + BN + ReLU |
| | Stage 6 | Conv 224–288 + BN + ReLU | | Conv 288–288 + BN + ReLU | | Conv 288–288 + BN + ReLU | | Dilation Conv(2) 288–288 + BN + ReLU | | Conv 288–288 + BN + ReLU | | Dilation Conv(2) 288–288 + BN + ReLU | | Conv 288–288 + BN + ReLU | | Dilation Conv(2) 288–288 + BN + ReLU | | Conv 288–288 + BN + ReLU | | Conv 288–288 + BN + ReLU |
| | Stage 7 | Conv 288–512 + BN + ReLU | | Conv 512–512 + BN + ReLU | | Conv 512–512 + BN + ReLU | | Dilation Conv(2) 512–512 + BN + ReLU | | Global average pooling | | Fully connected layer | | Softmax | | Classification result |
|
|
