| 168 | |
| 169 | class Discriminator(nn.Module): |
| 170 | def __init__(self, ngpu): |
| 171 | super(Discriminator, self).__init__() |
| 172 | self.ngpu = ngpu |
| 173 | self.main = nn.Sequential( |
| 174 | # input is (nc) x 64 x 64 |
| 175 | nn.Conv2d(nc, ndf, 4, 2, 1, bias=False), |
| 176 | nn.LeakyReLU(0.2, inplace=True), |
| 177 | # state size. (ndf) x 32 x 32 |
| 178 | nn.Conv2d(ndf, ndf * 2, 4, 2, 1, bias=False), |
| 179 | nn.BatchNorm2d(ndf * 2), |
| 180 | nn.LeakyReLU(0.2, inplace=True), |
| 181 | # state size. (ndf*2) x 16 x 16 |
| 182 | nn.Conv2d(ndf * 2, ndf * 4, 4, 2, 1, bias=False), |
| 183 | nn.BatchNorm2d(ndf * 4), |
| 184 | nn.LeakyReLU(0.2, inplace=True), |
| 185 | # state size. (ndf*4) x 8 x 8 |
| 186 | nn.Conv2d(ndf * 4, ndf * 8, 4, 2, 1, bias=False), |
| 187 | nn.BatchNorm2d(ndf * 8), |
| 188 | nn.LeakyReLU(0.2, inplace=True), |
| 189 | # state size. (ndf*8) x 4 x 4 |
| 190 | nn.Conv2d(ndf * 8, 1, 4, 1, 0, bias=False), |
| 191 | nn.Sigmoid() |
| 192 | ) |
| 193 | |
| 194 | def forward(self, input): |
| 195 | if (input.is_cuda or input.is_xpu) and self.ngpu > 1: |