Table 2. Layers and properties of ResNet64 architecture.
ResNet models have advantages in deep networks such as training simplicity, network augmentation, transfer learning capability, and backward adaptability.
| Layer | Output size | Filters | Activation | Params |
|---|---|---|---|---|
| Conv2d-1 | [batch_size, 64, 112, 112] | 64 | – | 9,408 |
| BatchNorm2d-2 | [batch_size, 64, 112, 112] | 64 | – | 128 |
| ReLU-3 | [batch_size, 64, 112, 112] | – | ReLU (inplace=True) | 0 |
| MaxPool2d-4 | [batch_size, 64, 56, 56] | – | – | 0 |
| Conv2d-5 | [batch_size, 64, 56, 56] | 64 | – | 4,096 |
| BatchNorm2d-6 | [batch_size, 64, 56, 56] | 64 | – | 128 |
| ReLU-7 | [batch_size, 64, 56, 56] | – | ReLU | 0 |
| Conv2d-8 | [batch_size, 64, 56, 56] | 64 | – | 36,864 |
| BatchNorm2d-9 | [batch_size, 64, 56, 56] | 64 | – | 128 |
| ReLU-10 | [batch_size, 64, 56, 56] | – | ReLU | 0 |
| Conv2d-11 | [batch_size, 256, 56, 56] | 256 | – | 16,384 |
| BatchNorm2d-12 | [batch_size, 256, 56, 56] | 256 | – | 512 |
| Conv2d-13 | [batch_size, 256, 56, 56] | 256 | – | 16,384 |
| BatchNorm2d-14 | [batch_size, 256, 56, 56] | 256 | – | 512 |
| ReLU-15 | [batch_size, 256, 56, 56] | – | ReLU | 0 |
| Bottleneck-16 | [batch_size, 256, 56, 56] | – | – | 0 |
| Conv2d-17 | [batch_size, 64, 56, 56] | 64 | – | 16,384 |
| BatchNorm2d-18 | [batch_size, 64, 56, 56] | 64 | – | 128 |
| ReLU-19 | [batch_size, 64, 56, 56] | – | ReLU | 0 |
| … | … | … | … | … |
| AdaptiveAvgPool2d-173 | [batch_size, 2048, 1, 1] | – | – | 0 |
| Linear-174 | [batch_size, 4] | 4 | – | Varies |
| ResNet-175 | [batch_size, 4] | – | – | 0 |