TABLE I. The Configuration Setting for Spatial-Channel-Attention ResNet.
| Layer name | Output size | Configuration |
|---|---|---|
| input |
 |
- |
| conv1 |
 |
 , 64, stride 2 |
 , max pool, stride 2 | ||
| conv2 |
 |
 |
| conv3 |
 |
 |
| conv4 |
 |
 |
| conv5 |
 |
 |
| hidden layer | 1024 | fc(1024,Relu) |
| hidden layer | 512 | fc(512,Relu) |
| hidden layer | 4 | fc(4,Relu) |
| output | 4 | softmax |
1 fc denotes fully-connected layer.
2 conv denotes convlution layer.
3 avg and max denote the average and max pooling operation respectively.
4 channel
and spatial
denote the channel-wise and spatial attention respectively.