Advances, Systems and Applications
Table 1 Overall architecture
From: Lightweight image classifier using dilated and depthwise separable convolutions
Type | Filter shape | Stride | Input size |
|---|---|---|---|
Dilated Conv | 3×3×32 | 1 | 224×224×3 |
Depthwise | 3×3×32 | 2 | 224×224×32 |
Separable Conv | 1×1×64 | 1 | 112×112×32 |
Dilated Conv | 3×3×64 | 1 | 112×112×64 |
Depthwise | 3×3×64 | 2 | 56×56×64 |
Separable Conv | 1×1×128 | 1 | 56×56×64 |
Dilated Conv | 3×3×128 | 1 | 1×1×128 |
Depthwise | 3×3×128 | 1 | 56×56×128 |
Separable Conv | 1×1×128 | 1 | 56×56×128 |
Dilated Conv | 3×3×128 | 1 | 56×56×128 |
Depthwise | 3×3×128 | 2 | 28×28×128 |
Separable Conv | 1×1×256 | 1 | 28×28×128 |
Dilated Conv | 3×3×256 | 1 | 28×28×256 |
Depthwise | 3×3×256 | 1 | 28×28×256 |
Separable Conv | 1×1×256 | 1 | 28×28×256 |
Dilated Conv | 3×3×256 | 1 | 28×28×256 |
Depthwise | 3×3×256 | 2 | 14×14×256 |
Separable Conv | 1×1×512 | 1 | 14×14×256 |
Dilated Conv | 3×3×512 | 1 | 14×14×512 |
Depthwise | 3×3×512 | 1 | 14×14×512 |
Separable Conv | 1×1×512 | 1 | 14×14×512 |
Dilated Conv | 3×3×512 | 1 | 14×14×512 |
Depthwise | 3×3×512 | 2 | 14×14×512 |
Separable Conv | 1×1×1024 | 1 | 7×7×512 |
Dilated Conv | 3×3×1024 | 1 | 7×7×1024 |
Depthwise | 3×3×1024 | 1 | 7×7×1024 |
Separable Conv | 1×1×1024 | 1 | 7×7×1024 |
Avg Pool | 7×7 | 1 | 7×7×1024 |
FC | 1024×class | 1 | 1×1×1024 |
softmax | Classifier | 1 | 1×1×class |