| 8 | |
| 9 | |
| 10 | class BasicBlock(nn.Module): |
| 11 | def __init__(self, c_in, c_out, is_downsample=False): |
| 12 | super(BasicBlock, self).__init__() |
| 13 | self.is_downsample = is_downsample |
| 14 | if is_downsample: |
| 15 | self.conv1 = nn.Conv2d( |
| 16 | c_in, c_out, 3, stride=2, padding=1, bias=False) |
| 17 | else: |
| 18 | self.conv1 = nn.Conv2d( |
| 19 | c_in, c_out, 3, stride=1, padding=1, bias=False) |
| 20 | self.bn1 = nn.BatchNorm2d(c_out) |
| 21 | self.relu = nn.ReLU(True) |
| 22 | self.conv2 = nn.Conv2d(c_out, c_out, 3, stride=1, |
| 23 | padding=1, bias=False) |
| 24 | self.bn2 = nn.BatchNorm2d(c_out) |
| 25 | if is_downsample: |
| 26 | self.downsample = nn.Sequential( |
| 27 | nn.Conv2d(c_in, c_out, 1, stride=2, bias=False), |
| 28 | nn.BatchNorm2d(c_out) |
| 29 | ) |
| 30 | elif c_in != c_out: |
| 31 | self.downsample = nn.Sequential( |
| 32 | nn.Conv2d(c_in, c_out, 1, stride=1, bias=False), |
| 33 | nn.BatchNorm2d(c_out) |
| 34 | ) |
| 35 | self.is_downsample = True |
| 36 | |
| 37 | def forward(self, x): |
| 38 | y = self.conv1(x) |
| 39 | y = self.bn1(y) |
| 40 | y = self.relu(y) |
| 41 | y = self.conv2(y) |
| 42 | y = self.bn2(y) |
| 43 | if self.is_downsample: |
| 44 | x = self.downsample(x) |
| 45 | return F.relu(x.add(y), True) |
| 46 | |
| 47 | |
| 48 | def make_layers(c_in, c_out, repeat_times, is_downsample=False): |