Advanced Deep Learning with Python

Depth-wise separable convolutions

An output slice in a cross-channel convolution receives input from all of the input slices using a single filter. The filter tries to learn features in a 3D space, where two of the dimensions are spatial (the height and width of the slice) and the third is the channel. Therefore, the filter maps both spatial and cross-channel correlations.

Depthwise separable convolutions (DSC, Xception: Deep Learning with Depthwise Separable Convolutions, https://arxiv.org/abs/1610.02357) can completely decouple cross-channel and spatial correlations. A DSC combines two operations: a depthwise convolution and a 1×1 convolution. In a depthwise convolution, a single input slice produces a single output slice, so it only maps ...