## Depthwise Separable Conv

• 操作的顺序。在TensorFlow等框架中，depthwise separable conv的实现是先使用channelwise的filter只在spatial dimension上做卷积，再使用$1\times 1$的卷积核做跨channel的融合。而Inception中先使用$1\times 1$的卷积核。
• 非线性变换的缺席。在Inception中，每个conv操作后面都有ReLU的非线性变换，而depthwise separable conv没有。

## Xception网络架构

we make the following hypothesis: that the mapping of cross-channels correlations and spatial correlations in the feature maps of convolutional neural networks can be entirely decoupled.

Xception的结构基于ResNet，但是将其中的卷积层换成了depthwise separable conv。如下图所示。整个网络被分为了三个部分：Entry，Middle和Exit。

The Xception architecture: the data first goes through the entry flow, then through the middle flow which is repeated eight times, and finally through the exit flow. Note that all Convolution and SeparableConvolution layers are followed by batch normalization [7] (not included in the diagram). All SeparableConvolution layers use a depth multiplier of 1 (no depth expansion).