Table 1.
Hyperparameters of the proposed architectures by custom design: k is the kernel size, s is stride, p is padding.
| Layer | No. of Filters or Neurons | k | s | p | Output Shape | Trainable Parameters |
|---|---|---|---|---|---|---|
| Input | – | – | – | – | (400, 400, 3) | 0 |
| block1_conv | 32 | 3 | 1 | same | (400, 400, 32) | 896 |
| block1_pool | 32 | 3 | 3 | 0 | (133, 133, 32) | 0 |
| block2_conv | 32 | 3 | 1 | same | (133, 133, 32) | 9248 |
| block2_pool | 32 | 2 | 2 | 0 | (66, 66, 32) | 0 |
| block3_conv | 64 | 3 | 1 | same | (66, 66, 64) | 18,496 |
| block3_pool | 64 | 2 | 2 | 0 | (33, 33, 64) | 0 |
| block4_conv | 64 | 3 | 1 | same | (33, 33, 64) | 36,928 |
| block4_pool | 64 | 2 | 2 | 0 | (16, 16, 64) | 0 |
| block5_conv | 64 | 3 | 1 | same | (16, 16, 64) | 36,928 |
| block5_pool | 64 | 2 | 2 | 0 | (8, 8, 64) | 0 |
| FC1 (architecture 1) | 1028 | – | – | – | – | 581,898,372 |
| FC1 (architecture 2) | 1028 | – | – | – | – | 143,296,004 |
| FC1 (architecture 3) | 1028 | – | – | – | – | 71,648,516 |
| FC1 (architecture 4) | 1028 | – | – | – | – | 16,843,780 |
| FC1 (architecture 5) | 1028 | – | – | – | – | 4,211,716 |
| FC2 | 2 | – | – | – | – | 2058 |
| Architecture 1 | – | – | – | – | – | 581,901,326 |
| Architecture 2 | – | – | – | – | – | 143,308,206 |
| Architecture 3 | – | – | – | – | – | 71,679,214 |
| Architecture 4 | – | – | – | – | – | 16,911,406 |
| Architecture 5 | – | – | – | – | – | 4,316,270 |