Method __init__

lib/Modules.py:105–137

(self, dilation_series=[6, 12, 18], padding_series=[6, 12, 18], depth=128)

# 目前太肥了，这玩意儿输出一个12*12的图片，需要2048的channel？
class GPM(nn.Module):
    def __init__(self, dilation_series=[6, 12, 18], padding_series=[6, 12, 18], depth=128):
        # def __init__(self, dilation_series=[2, 5, 7], padding_series=[2, 5, 7], depth=128):
        super(GPM, self).__init__()
        self.branch_main = nn.Sequential(
            nn.AdaptiveAvgPool2d((1, 1)),
            BasicConv2d(2048, depth, kernel_size=1, stride=1)
        )
        self.branch0 = BasicConv2d(2048, depth, kernel_size=1, stride=1)
        self.branch1 = BasicConv2d(2048, depth, kernel_size=3, stride=1, padding=padding_series[0],
                                   dilation=dilation_series[0])
        self.branch2 = BasicConv2d(2048, depth, kernel_size=3, stride=1, padding=padding_series[1],
                                   dilation=dilation_series[1])
        self.branch3 = BasicConv2d(2048, depth, kernel_size=3, stride=1, padding=padding_series[2],
                                   dilation=dilation_series[2])
        self.head = nn.Sequential(
            BasicConv2d(depth * 5, 256, kernel_size=3, padding=1),
            PAM(256)
        )
        self.out = nn.Sequential(
            nn.Conv2d(256, 64, 3, padding=1),
            nn.BatchNorm2d(64, affine=affine_par),
            nn.PReLU(),
            nn.Dropout2d(p=0.1),
            nn.Conv2d(64, 1, 1)
        )

        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
                m.weight.data.normal_(0, 0.01)
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

    def forward(self, x):
        # out = self.conv2d_list[0](x)