def forward()

in src/controlnet_aux/pidi/model.py [0:0]

• 43 lines of code
• 9 McCabe index (conditional complexity)

    def forward(self, x):
        H, W = x.size()[2:]

        x = self.init_block(x)

        x1 = self.block1_1(x)
        x1 = self.block1_2(x1)
        x1 = self.block1_3(x1)

        x2 = self.block2_1(x1)
        x2 = self.block2_2(x2)
        x2 = self.block2_3(x2)
        x2 = self.block2_4(x2)

        x3 = self.block3_1(x2)
        x3 = self.block3_2(x3)
        x3 = self.block3_3(x3)
        x3 = self.block3_4(x3)

        x4 = self.block4_1(x3)
        x4 = self.block4_2(x4)
        x4 = self.block4_3(x4)
        x4 = self.block4_4(x4)

        x_fuses = []
        if self.sa and self.dil is not None:
            for i, xi in enumerate([x1, x2, x3, x4]):
                x_fuses.append(self.attentions[i](self.dilations[i](xi)))
        elif self.sa:
            for i, xi in enumerate([x1, x2, x3, x4]):
                x_fuses.append(self.attentions[i](xi))
        elif self.dil is not None:
            for i, xi in enumerate([x1, x2, x3, x4]):
                x_fuses.append(self.dilations[i](xi))
        else:
            x_fuses = [x1, x2, x3, x4]

        e1 = self.conv_reduces[0](x_fuses[0])
        e1 = F.interpolate(e1, (H, W), mode="bilinear", align_corners=False)

        e2 = self.conv_reduces[1](x_fuses[1])
        e2 = F.interpolate(e2, (H, W), mode="bilinear", align_corners=False)

        e3 = self.conv_reduces[2](x_fuses[2])
        e3 = F.interpolate(e3, (H, W), mode="bilinear", align_corners=False)

        e4 = self.conv_reduces[3](x_fuses[3])
        e4 = F.interpolate(e4, (H, W), mode="bilinear", align_corners=False)

        outputs = [e1, e2, e3, e4]

        output = self.classifier(torch.cat(outputs, dim=1))
        #if not self.training:
        #    return torch.sigmoid(output)

        outputs.append(output)
        outputs = [torch.sigmoid(r) for r in outputs]
        return outputs