## Depthwise Separable Convolution - How does it works?

I notice that both EfficientPS and Panoptic-DeepLab architectures uses depthwise separable convolution to reduce the total number of parameters, as a result, speed up convolution. This is a form of factorized convolutions which factorize a standard convolution into a depthwise convolution and a 1 x 1 convolution called a pointwise convolution. Depthwise separable convolution first use the depthwise convolution to applies a single filter to each input channel. Then the pointwise convolution is used to applies a 1 x 1 convolution to combine the outputs of the depthwise convolution....