TODO

+1、近端梯度下降法
https://zhuanlan.zhihu.com/p/82622940
+2、nn.init.kaiming_uniform_使用，在conv后面,类似BN层？

 conv = Conv2d(
            in_channels, 
            out_channels, 
            kernel_size=kernel_size, 
            stride=stride, 
            padding=dilation * (kernel_size - 1) // 2, 
            dilation=dilation, 
            bias=False if use_gn else True
        )
        # Caffe2 implementation uses XavierFill, which in fact
        # corresponds to kaiming_uniform_ in PyTorch
        nn.init.kaiming_uniform_(conv.weight, a=1)