Deep Learning with PyTorch Quick Start Guide

Implementing a pretrained model

Remember the Guiseppe toys dataset we played with in Chapter 1, Introduction to PyTorch? We now finally have the tools and knowledge to be able to create a classification model for this data. We are going to do this by using a model pretrained on the Imagenet dataset. This is called transfer learning, because we are transferring the learning achieved on one dataset to make predictions on a different, usually much smaller, dataset. Using a network with pretrained weights dramatically increases its performance on much smaller datasets, and this is surprisingly easy to achieve. In the simplest case, we can pass the pretrained model a data of labeled images and simply change the number of output features. Remember ...