Howard and others (https://arxiv.org/pdf/1704.04861.pdf) introduced a new class of models called MobileNets that can be used for mobile and embedded applications. MobileNets can be used for different applications such as object detection, landmark recognition, face attributes, fine-grain classification as shown here:
Reproduced from Howard and others
MobileNets reduced the size and computation of models by replacing standard convolution filters (a) with depthwise (b) and pointwise convolutions (c) as shown here:
Reproduced from ...
