Pyramid Stereo Matching Network:SPP module

submodule.py

class feature_extraction(nn.Module):

 

  self.branch1 = nn.Sequential(nn.AvgPool2d((64, 64), stride=(64,64)),
                                     convbn(128, 32, 1, 1, 0, 1),
                                     nn.ReLU(inplace=True))

        self.branch2 = nn.Sequential(nn.AvgPool2d((32, 32), stride=(32,32)),
                                     convbn(128, 32, 1, 1, 0, 1),
                                     nn.ReLU(inplace=True))

        self.branch3 = nn.Sequential(nn.AvgPool2d((16, 16), stride=(16,16)),
                                     convbn(128, 32, 1, 1, 0, 1),
                                     nn.ReLU(inplace=True))

        self.branch4 = nn.Sequential(nn.AvgPool2d((8, 8), stride=(8,8)),
                                     convbn(128, 32, 1, 1, 0, 1),
                                     nn.ReLU(inplace=True))

        self.lastconv = nn.Sequential(convbn(320, 128, 3, 1, 1, 1),
                                      nn.ReLU(inplace=True),
                                      nn.Conv2d(128, 32, kernel_size=1, padding=0, stride = 1, bias=False))

对应下面结构。

submodule.py

class feature_extraction(nn.Module):

def forward(self, x):
        print('feature_extraction forward 123')
        output      = self.firstconv(x)
        output      = self.layer1(output)
        output_raw  = self.layer2(output)
        output      = self.layer3(output_raw)
        output_skip = self.layer4(output)

        output_branch1 = self.branch1(output_skip)
        output_branch1 = F.upsample(output_branch1, (output_skip.size()[2],output_skip.size()[3]),mode='bilinear')

        output_branch2 = self.branch2(output_skip)
        output_branch2 = F.upsample(output_branch2, (output_skip.size()[2],output_skip.size()[3]),mode='bilinear')

        output_branch3 = self.branch3(output_skip)
        output_branch3 = F.upsample(output_branch3, (output_skip.size()[2],output_skip.size()[3]),mode='bilinear')

        output_branch4 = self.branch4(output_skip)
        output_branch4 = F.upsample(output_branch4, (output_skip.size()[2],output_skip.size()[3]),mode='bilinear')

        output_feature = torch.cat((output_raw, output_skip, output_branch4, output_branch3, output_branch2, output_branch1), 1)
        output_feature = self.lastconv(output_feature)

        return output_feature