Table 1.
Cardio-Net with feature concatenation, where the Downsample and Upsample blocks include Convolution (Conv), Bottleneck convolution (Bottleneck-C), Depth-wise separable convolution (DW-Sep-Conv), Concatenation, and Pool. Batch normalization and ReLU layers are used with convolutions and denoted as “**”.
| Block | Layer Name | Layer Size (Height × Width × Number of Channels), (Stride) |
Filters/Groups | Output |
|---|---|---|---|---|
| Downsample block |
Conv-1-1 ** | 3 × 3 × 64 (S = 1) | 64 | 350 × 350 × 64 |
| Conv-1-2 | 3 × 3 × 64 (S = 1) | 64 | 350 × 350 × 64 | |
| Concatenation-1 | 350 × 350 × 128 | |||
| Bottleneck-C-1 ** | 1 × 1 (S = 1) | 64 | 350 × 350 × 64 | |
| Pool-1 2 × 2 (S = 2) | 175 × 175 × 64 | |||
| Conv-2-1 ** | 3 × 3 × 64 (S = 1) | 128 | 175 × 175 × 128 | |
| Conv-2-2 | 3 × 3 × 128 (S = 1) | 128 | 175 × 175 × 128 | |
| Concatenation-2 | 175 × 175 × 256 | |||
| Bottleneck-C-2 ** | 1 × 1 (S = 1) | 128 | 175 × 175 × 128 | |
| Pool-2 2 × 2 (S = 2) | 87 × 87 × 128 | |||
| Conv-3-1 ** | 3 × 3 × 128 (S = 1) | 256 | 87 × 87 × 256 | |
| DW-Sep-Conv-3-2 | 3 × 3 × 256 (S = 1) | 256 | 87 × 87 × 256 | |
| Concatenation-3 | 87 × 87 × 512 | |||
| Bottleneck-C-3 ** | 1 × 1 (S = 1) | 128 | 87 × 87 × 256 | |
| Pool-3 2 × 2 (S = 2) | 43 × 43 × 256 | |||
| DW-Sep-Conv-4-1 ** | 3 × 3 × 256 (S = 1) | 256 | 43 × 43 × 256 | |
| DW-Sep-Conv-4-2 | 3 × 3 × 256 (S = 1) | 256 | 43 × 43 × 256 | |
| Concatenation-4 | 43 × 43 × 512 | |||
| Bottleneck-C-4 ** | 1 × 1 (S = 1) | 256 | 43 × 43 × 256 | |
| Pool-4 2 × 2 (S = 2) | 21 × 21 × 256 | |||
| Upsample block |
UnPool-4 2 × 2 (S = 2) | 43 × 43 × 256 | ||
| DW-Sep-Conv-4-2 ** | 3 × 3 × 256 (S = 1) | 256 | 43 × 43 × 256 | |
| DW-Sep-Conv-4-1 | 3 × 3 × 256 (S = 1) | 256 | 43 × 43 × 256 | |
| Concatenation-5 | 43 × 43 × 512 | |||
| Bottleneck-C-5 ** | 1 × 1 (S = 1) | 256 | 43 × 43 × 256 | |
| UnPool-3 2 × 2 (S = 2) | 87 × 87 × 256 | |||
| DW-Sep-Conv-3-2 ** | 3 × 3 × 256 (S = 1) | 256 | 87 × 87 × 256 | |
| Conv-3-1 | 3 × 3 × 256 (S = 1) | 128 | 87 × 87 × 128 | |
| Concatenation-6 | 87 × 87 × 640 | |||
| Bottleneck-C-6 ** | 1 × 1 (S = 1) | 128 | 87 × 87 × 128 | |
| UnPool-2 2 × 2 (S = 2) | 175 × 175 × 128 | |||
| Conv-2-2 ** | 3 × 3 × 128 (S = 1) | 128 | 175 × 175 × 128 | |
| Conv-2-1 | 3 × 3 × 128 (S = 1) | 64 | 175 × 175 × 64 | |
| Concatenation-7 | 175 × 175 × 320 | |||
| Bottleneck-C-7 ** | 1 × 1 (S = 1) | 64 | 175 × 175 × 64 | |
| UnPool-1 2 × 2 (S = 2) | 350 × 350 × 64 | |||
| Conv-1-2 ** | 3 × 3 × 64 (S = 1) | 64 | 350 × 350 × 64 | |
| Conv-1-1 | 3 × 3 × 64 (S = 1) | 64 | 350 × 350 × 64 | |
| Concatenation-8 | 350 × 350 × 160 | |||
| Bottleneck-C-8 ** | 1 × 1 (S = 1) | 2 | 350 × 350 × 2 | |