Table 2.
The 2D classifier model architecture.
| Layer | Output Dimensions |
|---|---|
| Input | [128, 128, 3] |
| Convolution | [128, 128, 8] |
| Convolution | [128, 128, 16] |
| Max Pool | [16, 64] |
| Convolution | [16, 64] |
| Convolution | [32, 64] |
| Max Pool | [32] |
| Convolution | [32] |
| Convolution | [32, 64] |
| Max Pool | [16, 64] |
| Convolution | [16, 64] |
| Convolution | [16, 16, 128] |
| Max Pool | [8, 8, 128] |
| Convolution | [8, 8, 128] |
| Convolution | [8, 8, 256] |
| Average Pool | [4, 4, 256] |
| Flatten | 4096 |
| Dense (relu) | 100 |
| Dropout (0.3) | 100 |
| Dense (softmax) | 43 |
All convolutional layers use 3 × 3 kernels and are followed by both ReLU activation and batch normalization.