Setting up the Transfer Learning model
The current recipe will cover Transfer Learning using the CIFAR-10 dataset. The previous recipe presented how to use a pretrained model. The current recipe will demonstrate how to use a pretrained model for different problem statements.
We will use another very good deep learning package, MXNET, to demonstrate the concept with another architecture, Inception. To simplify the computation, we will reduce the problem complexity from 10 classes to two classes: aeroplane and automobile. The recipe focuses on data preparation for Transfer Learning using Inception-BN.
Getting ready
The section prepares for the upcoming section for setting-up Transfer Learning model.
- Download the CIFAR-10 dataset from http://www.cs.toronto.edu/~kriz/cifar.html. The
download.cifar.data
function from Chapter 3,Convolution Neural Networks, can be used to download the dataset. - Install the
imager
package:
install.packages("imager")
How to do it...
The current part of the recipe will provide...