在?https://blog.csdn.net/zhqh100/article/details/90376732?中介紹MobileNet時，提到?DepthwiseConv2D，這是keras中的API，那mxnet中該 API 叫什么名字呢？

我跟蹤了一下代碼，以及打印summary之類的，基本搞清楚了，mxnet 中無論是?DepthwiseConv2D 還是?Conv2D，統一都叫做?Conv2D，參考如下調用代碼（/home/luke/miniconda3/lib/python3.6/site-packages/gluoncv/model_zoo/mobilenet.py）：

# pylint: disable= too-many-arguments
def _add_conv(out, channels=1, kernel=1, stride=1, pad=0,num_group=1, active=True, relu6=False, norm_layer=BatchNorm, norm_kwargs=None):out.add(nn.Conv2D(channels, kernel, stride, pad, groups=num_group, use_bias=False))print("kernel = %d, groups = %d, channels = %d" % (kernel, num_group, channels))out.add(norm_layer(scale=True, **({} if norm_kwargs is None else norm_kwargs)))if active:out.add(RELU6() if relu6 else nn.Activation('relu'))

區別在于其中的?groups 參數，http://mxnet.incubator.apache.org/api/python/gluon/nn.html

• groups?(int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.

而這里在調用時，如果需要分離卷積的話，channels值和groups值相等就可以了

            _add_conv(self.out,in_channels * t,kernel=3,stride=stride,pad=1,num_group=in_channels * t,relu6=True,norm_layer=BatchNorm, norm_kwargs=None)

值都為?in_channels * t。

總結

以上是生活随笔為你收集整理的mxnet 中的 DepthwiseConv2D API的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。