In this article, I’m going to walk through the architectural implementations that led from the first MobileNet architecture to the state of the art EfficientNet architecture, with a bunch of improvements in between. The MobileNet architecture was aimed at reducing the computational requirements for large neural networks without a large drop in performance on the classification task. This was with the particular goal of deploying the networks on edge devices, like mobile phones. After several years of iterat

Source: ttumiel