Table 2 Generator architecture: we adopted the input noise vector of dimensionality 100 drawn from a zero-mean Gaussian distribution.
| Â | Input projection | Layer1 | Layer2 | Layer3 | Layer4 | Layer5 | Layer6 |
|---|---|---|---|---|---|---|---|
Type | Fully connected | Fractionally strided convolution | Fractionally strided convolution | Fractionally strided convolution | Fractionally strided convolution | Fractionally strided convolution | Fractionally strided convolution |
Input | [1 × 100] | [4 × 4 × 1024] | [8 × 8 × 512] | [16 × 16 × 256] | [32 × 32 × 128] | [64 × 64 × 64] | [128 × 128 × 32] |
Output | [4 × 4 × 1024] | [8 × 8 × 512] | [16 × 16 × 256] | [32 × 32 × 128] | [64 × 64 × 64] | [128 × 128 × 32] | [64 × 64 × 2] |
Activation | ReLU | ReLU | ReLU | ReLU | ReLU | ReLU | TanH |
Batch norm | Yes | Yes | Yes | Yes | Yes | Yes | Yes |
Stride | – | 2 | 2 | 2 | 2 | 1 | – |
Padding | – | same | Same | Same | Same | Same | Same |
Kernel Size | – | 5 | 5 | 5 | 5 | 5 | 5 |
Kernels | – | 1024 | 512 | 256 | 128 | 64 | 32 |
