Introduction

During the last couple of years, a lot of game-changing network architectures have been proposed and published. Most of them open-sourced their code or published their weights. If the latter was not the case, others implemented the network architecture and shared the weights. As a result, many deep learning frameworks give direct access to popular models and their weights. In this chapter, we will demonstrate how to leverage these pretrained weights. Most of these models have been trained on large image datasets used in competitions, such as the ImageNet dataset. This dataset has been published for the ImageNet Large Scale Visual Recognition Challenge (ILSVRC). By leveraging these pretrained weights, we can obtain good results and reduce training time.