Table 1.
Parameters distributed in HDM-NET with classifiers (SVM), dw means depthwise.
| Layer Type | Stride | Kernel | Feature Map |
|---|---|---|---|
| Conv. Layer | 2 | 3 × 3 | 3 × 32 |
| Depthwise Conv. | 1 | 3 × 3 | 32 dw |
| Conv. Layer | 1 | 1 × 1 | 32 × 64 |
| Depthwise Conv. | 2 | 3 × 3 | 64 dw |
| Conv. Layer | 1 | 1 × 1 | 64 × 128 |
| Depthwise Conv. | 1 | 3 × 3 | 128 dw |
| Conv. Layer | 1 | 1 × 1 | 128 × 128 |
| Depthwise Conv. | 2 | 3 × 3 | 128 dw |
| Conv. Layer | 1 | 1 × 1 | 128 × 256 |
| Depthwise Conv. | 1 | 3 × 3 | 256 dw |
| Conv. Layer | 1 | 1 × 1 | 256 × 256 |
| Depthwise Conv. | 2 | 3 × 3 | 256 dw |
| Conv. Layer | 1 | 1 × 1 | 256 × 512 |
| 2×Depthwise | 1 | 3 × 3 | 512 dw |
| 2×Conv. Layer | 1 | 1 × 1 | 512 × 512 |
| Depthwise Conv. | 2 | 3 × 3 | 512 dw |
| Conv. Layer | 1 | 1 × 1 | 512 × 1024 |
| Depthwise Conv. | 1 | 3 × 3 | 1024 dw |
| Conv. Layer | 1 | 1 × 1 | 1024 × 1024 |
| Global AvgPool | 1 | 7×7 | 1024 × 1024 |
| BN | --- | --- | 1024 × 1024 |
| ReLU | --- | --- | 1024 × 1024 |
| Classifier (SVM) | --- | --- | 3 |