Research Article
A Generative Image Inpainting Model Based on Edge and Feature Self-Arrangement Constraints
Table 3
Network parameters of the image generator.
| | Layer | Inputs (H × W × C) | Kernel size | Stride | Padding | Dilation | Activation function | Outputs (H × W × C) |
| | Conv | 256 × 256 × 4 | 4 × 4 | 2 | 1 | 1 | — | 128 × 128 × 64 | | Conv | 128 × 128 × 64 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 64 × 64 × 128 | | Conv | 64 × 64 × 128 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 32 × 32 × 256 | | Conv | 32 × 32 × 256 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 16 × 16 × 512 | | Conv | 16 × 16 × 512 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 8 × 8 × 512 | | Conv | 8 × 8 × 512 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 4 × 4 × 512 | | Conv | 4 × 4 × 512 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 2 × 2 × 512 | | Conv | 2 × 2 × 512 | 4 × 4 | 2 | 1 | 1 | Leaky ReLU | 1 × 1 × 512 | | DeConv | 1 × 1 × 512 | 4 × 4 | 2 | 1 | 1 | ReLU | 2 × 2 × 512 | | DeConv | 2 × 2 × 1024 | 4 × 4 | 2 | 1 | 1 | ReLU | 4 × 4 × 512 | | DeConv | 4 × 4 × 1024 | 4 × 4 | 2 | 1 | 1 | ReLU | 8 × 8 × 512 | | DeConv | 8 × 8 × 1024 | 4 × 4 | 2 | 1 | 1 | ReLU | 16 × 16 × 512 | | DeConv | 16 × 16 × 1024 | 4 × 4 | 2 | 1 | 1 | ReLU | 32 × 32 × 256 | | DeConv | 32 × 32 × 768 | 4 × 4 | 2 | 1 | 1 | ReLU | 64 × 64 × 128 | | DeConv | 64 × 64 × 256 | 4 × 4 | 2 | 1 | 1 | ReLU | 128 × 128 × 64 | | DeConv | 128 × 128 × 128 | 4 × 4 | 2 | 1 | 1 | Tanh | 256 × 256 × 3 |
|
|
