refernence paper:
《The One Hundred Layers Tiramisu: Fully Convolutional DenseNets for Semantic Segmentation》
Simon Jégou, Michal Drozdzal, David Vazquez, Adriana Romero, Yoshua Bengio
https://arxiv.org/abs/1611.09326
input size:512x512
use 7x7 conv and 2x2 stride down sample to 256x256
use 2x2 maxpooling down sample to 128x128
output size:128x128，interpolation—>512x512
Dense block groth rate:24
| Layer | feature map | filter | Conv | DB size |
|---|---|---|---|---|
| Input | 512x512 | 1 | NULL | |
| Conv0 | 256x256 | 48 | 7x7 | |
| Maxpooling | 128x128 | 48 | 2x2 | |
| DB1 | 128x128 | 120 | 3x3 | 3 |
| TD1 | 64x64 | 120 | 1x1 | |
| DB2 | 64x64 | 240 | 3x3 | 5 |
| TD2 | 32x32 | 240 | 1x1 | |
| DB3 | 32x32 | 456 | 3x3 | 7 |
| TD3 | 16x16 | 456 | 1x1 | |
| Center | 16x16 | 672 | 3x3 | 9 |
| TU5 | 32x32 | 456 | 2x2 | |
| C5(TU5,DB3) | 32x32 | 912 | NULL | |
| DB5 | 32x32 | 456 | 3x3 | 7 |
| TU6 | 64x64 | 240 | 2x2 | |
| C6(TU6,DB2) | 64x64 | 480 | NULL | |
| DB6 | 64x64 | 240 | 3x3 | 5 |
| TU7 | 128x128 | 120 | 2x2 | |
| C7(TU7,DB1) | 128x128 | 240 | NULL | |
| DB7 | 128x128 | 120 | 3x3 | 3 |
| Output | 128x128 | 1 | 1x1 |
layer:51
total size of network:26Mb