Table 1 Detailed parameters of 3D Res U-Net
From: Design of lung nodules segmentation and recognition algorithm based on deep learning
Name | Operations | Output size |
|---|---|---|
Input | 1 × 48 × 192 × 192 | |
Encoder0 | Conv, IN, ReLU, c = 8, k = 1, p = 0 | 8 × 24 × 96 × 96 |
Conv, IN, ReLU, c = 8, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 8, k = 3, p = 1 | ||
MaxPool, k = 2 | ||
Encoder1 | Conv, IN, ReLU, c = 16, k = 1, p = 0 | 16 × 12 × 48 × 48 |
Conv, IN, ReLU, c = 16, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 16, k = 3, p = 1 | ||
MaxPool, k = 2 | ||
Encoder2 | Conv, IN, ReLU, c = 32, k = 1, p = 0 | 32 × 6 × 24 × 24 |
Conv, IN, ReLU, c = 32, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 32, k = 3, p = 1 | ||
MaxPool, k = 2 | ||
Encoder3 | Conv, IN, ReLU, c = 64, k = 1, p = 0 | 64 × 3 × 12 × 12 |
Conv, IN, ReLU, c = 64, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 64, k = 3, p = 1 | ||
MaxPool, k = 2 | ||
Bottle | Conv, IN, ReLU, c = 128, k = 1, p = 0 | 128 × 3 × 12 × 12 |
Conv, IN, ReLU, c = 128, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 128, k = 3, p = 1 | ||
Decoder0 | TransConv, IN, ReLU, c = 64, k = 2, s = 2 | 64 × 6 × 24 × 24 |
Conv, IN, ReLU, c = 64, k = 1, p = 0 | ||
Conv, IN, ReLU, c = 64, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 64, k = 3, p = 1 | ||
Decoder1 | TransConv, IN, ReLU, c = 32, k = 2, s = 2 | 32 × 12 × 48 × 48 |
Conv, IN, ReLU, c = 32, k = 1, p = 0 | ||
Conv, IN, ReLU, c = 32, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 32, k = 3, p = 1 | ||
Decoder2 | TransConv, IN, ReLU, c = 16, k = 2, s = 2 | 16 × 24 × 96 × 96 |
Conv, IN, ReLU, c = 16, k = 1, p = 0 | ||
Conv, IN, ReLU, c = 16, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 16, k = 3, p = 1 | ||
Decoder3 | TransConv, IN, ReLU, c = 8, k = 2, s = 2 | 8 × 48 × 192 × 192 |
Conv, IN, ReLU, c = 8, k = 1, p = 0 | ||
Conv, IN, ReLU, c = 8, k = 3, p = 1 | ||
Conv, IN, ReLU, c = 8, k = 3, p = 1 | ||
OutputConv | Conv, Sigmoid, c = 1, k = 1, p = 0 | 1 × 48 × 192 × 192 |