Table B 1.
Network parameter
| Layer | Output volume | Output feature size | Kernel Size | Stride |
|---|---|---|---|---|
| First convolution | 128×128×48 | 48 | 3×3×3 | (1,1,1) |
| Dense block (3 layers) | 128×128×48 | 48+48a | 3×3×3 | (1,1,1) |
| Transition down | 64×64×48 | 96 | 1×1×1 | (2,2,1) |
| Dense block (4 layers) | 64×64×48 | 64+96 | 3×3×3 | (1,1,1) |
| Transition down | 32×32×48 | 160 | 1×1×1 | (2,2,1) |
| Dense block (5 layers) | 32×32×48 | 80+160 | 3×3×3 | (1,1,1) |
| Transition down | 16×16×48 | 240 | 1×1×1 | (2,2,1) |
| Dense block (6 layers) | 16×16×48 | 96+240 | 3×3×3 | (1,1,1) |
| Transition down | 8×8×48 | 336 | 1×1×1 | (2,2,1) |
| bottle neck (7 layers) | 8×8×48 | 112 | 3×3×3 | (1,1,1) |
| Transition up | 16×16×48 | 112+336 | 1×1×1 | (2,2,1) |
| Dense block (6 layers) | 16×16×48 | 96 | 3×3×3 | (1,1,1) |
| Transition up | 32×32×48 | 96+240 | 1×1×1 | (2,2,1) |
| Dense block (5 layers) | 32×32×48 | 80 | 3×3×3 | (1,1,1) |
| Transition up | 64×64×48 | 80+160 | 1×1×1 | (2,2,1) |
| Dense block (4 layers) | 64×64×48 | 64 | 3×3×3 | (1,1,1) |
| Transition up | 128×128×48 | 64+96 | 1×1×1 | (2,2,1) |
| Dense block (3 layers) | 128×128×48 | 48 | 3×3×3 | (1,1,1) |
| Last convolution | 128×128×48 | 2 | 3×3×3 | (1,1,1) |
a
+ devotes for skip connection from previous layer output or corresponding dense block