Skip to main content

Table 2 The details of the discriminative network D

From: Compression artifacts reduction by improved generative adversarial networks

Discriminative network D

Input: G(x) or ground truth, size = 64 × 64

[Conv 1]

Conv (4, 4, 64), stride = 2, padding = 1; BatchNorm; LeakyReLU;

[Conv 2]

Conv (4, 4, 128), stride = 2, padding = 1; BatchNorm; LeakyReLU;

[Conv 3]

Conv (4, 4, 256), stride = 2, padding = 1; BatchNorm; LeakyReLU;

[Conv 4]

Conv (4, 4, 512), stride = 2, padding = 1; BatchNorm; LeakyReLU;

[Conv 5]

Conv (4, 4, 1), stride = 1, padding = 0; Sigmoid

Output: size = 1 × 1