TABLE 1.
Comparison of the VGG‐16 and VGG‐19 with the LBTS‐Net16 and LBTS‐Net19 architectures in terms of accuracy and complexity where “DW‐Conv” represents the depth‐wise convolution operation
| Layer | VGG‐19 | LBTS‐Net19 | VGG‐16 | LBTS‐Net16 |
|---|---|---|---|---|
| Layer 1 | Conv,64 | Conv,32 | Conv,64 | Conv,32 |
| Layer 2 | Conv,64 | DW Conv,32 | Conv,64 | DW Conv,32 |
| Layer 3 | Conv,128 | DW Conv,64 | Conv,128 | DW Conv,64 |
| Layer 4 | Conv,128 | DW Conv,64 | Conv,128 | DW Conv,64 |
| Layer 5 | Conv,256 | Conv,256 | Conv,256 | Conv,256 |
| Layer 6 | Conv,256 | Conv,256 | Conv,256 | Conv,256 |
| Layer 7 | Conv,256 | Conv,256 | Conv,256 | Conv,256 |
| Layer 8 | Conv,256 | Conv,256 | Conv,512 | Conv,512 |
| Layer 9 | Conv,512 | Conv,512 | Conv,512 | Conv,512 |
| Layer 10 | Conv,512 | Conv,512 | Conv,512 | Conv,512 |
| Layer 11 | Conv,512 | Conv,512 | Conv,512 | Conv,512 |
| Layer 12 | Conv,512 | Conv,512 | Conv,512 | Conv,512 |
| Layer 13 | Conv,512 | Conv,512 | Conv,512 | Conv,512 |
| Layer 14 | Conv,512 | Conv,512 | Fc,4096 | Fc,4096 |
| Layer 15 | Conv,512 | Conv,512 | Fc,4096 | Fc,4096 |
| Layer 16 | Conv,512 | Conv,512 | output layer with softmax | output layer with softmax |
| Layer 17 | Fc,4096 | Fc,4096 | – | – |
| Layer 18 | Fc,4096 | Fc,4096 | – | – |
| Layer 19 | output layer with softmax | output layer with softmax | – | – |
| # of parameters | 1.43 × 108 | 7 × 107 | 1.38 × 108 | 6.5 × 107 |
| Memory | 549 Mb | 269 Mb | 528 Mb | 248 Mb |
| Accuracy | 98.66% | 98.32% | 98.46% | 98.11% |