(self, x)
| 50 | self.dropblock = DropBlock2D(dropblock_prob, 3) |
| 51 | |
| 52 | def forward(self, x): |
| 53 | x = self.conv(x) |
| 54 | if self.use_bn: |
| 55 | x = self.bn0(x) |
| 56 | if self.dropblock_prob > 0.0: |
| 57 | x = self.dropblock(x) |
| 58 | x = self.relu(x) |
| 59 | |
| 60 | batch, channel = x.shape[:2] |
| 61 | if self.radix > 1: |
| 62 | splited = torch.split(x, channel//self.radix, dim=1) |
| 63 | gap = sum(splited) |
| 64 | else: |
| 65 | gap = x |
| 66 | gap = F.adaptive_avg_pool2d(gap, 1) |
| 67 | gap = self.fc1(gap) |
| 68 | |
| 69 | if self.use_bn: |
| 70 | gap = self.bn1(gap) |
| 71 | gap = self.relu(gap) |
| 72 | |
| 73 | atten = self.fc2(gap).view((batch, self.radix, self.channels)) |
| 74 | if self.radix > 1: |
| 75 | atten = F.softmax(atten, dim=1).view(batch, -1, 1, 1) |
| 76 | else: |
| 77 | atten = F.sigmoid(atten, dim=1).view(batch, -1, 1, 1) |
| 78 | |
| 79 | if self.radix > 1: |
| 80 | atten = torch.split(atten, channel//self.radix, dim=1) |
| 81 | out = sum([att*split for (att, split) in zip(atten, splited)]) |
| 82 | else: |
| 83 | out = atten * x |
| 84 | return out.contiguous() |
| 85 | |
| 86 | |
| 87 | class GlobalAvgPool2d(nn.Module): |
nothing calls this directly
no outgoing calls
no test coverage detected