Table 3 DeepScore architecture for 2 x 512 x 512 inputs
From: DeepFocus: fast focus and astigmatism correction for electron microscopy
Layer # | Layer Type | Input Channels | Output Channels | Kernel Size | Pooling Size | Activation | Dropout (p) | BatchNorm. |
|---|---|---|---|---|---|---|---|---|
1 | Conv3D | 1 | 20 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
2 | Conv3D | 20 | 30 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
3 | Conv3D | 30 | 40 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
4 | Conv3D | 40 | 50 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
5 | Conv3D | 50 | 60 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
6 | Conv3D | 60 | 70 | (1, 3, 3) | (1, 2, 2) | ReLU | 0.1 | Yes |
7 | Linear | 2520 | 250 | - | - | ReLU | - | No |
8 | Linear | 250 | 50 | - | - | ReLU | - | No |
9 | Linear | 50 | 2 | - | - | - | - | No |
