MobileNet: Paper Review and Model Architecture

Introduction

Papers are always long and full of details. To extract the key concepts and to convert the knowledges to my own language for a quicker review in the future, I drew mind-maps for some papers I read. There are many model architectures in the original papers or other posts, to have a “better” illustration, I drew partial of each with Google Drawings by myself. This post shows the light weight model — MobileNet.

Depth-wise Separable Convolutions Architecture

Depthwise Separabel Convolutions Architecture and params, flops calculation

Link for the above Google Drawings: https://docs.google.com/drawings/d/1mADAQA_YGEl-LBH8hYcW2dPDU_nuFLVuD8wAtbdjuQI/edit?usp=sharing

Mind Map

A part of MobileNet’s mind map

Link for MobileNet’s mind map: http://www.xmind.net/m/LM8u