Image classification and transfer learning
So far, we have discussed what image classification is all about. In this section, we will get our hands dirty by building our own classifiers. In one of the earlier sections of the chapter, we briefly mentioned famous benchmarking datasets, including CIFAR-10 and Stanford Dogs datasets, which we will be concentrating on in the coming sections. We will also utilize pretrained models to understand how we can leverage transfer learning to improve upon our models.
CIFAR-10
CIFAR-10 is one of the most widely used image datasets in the world of deep learning. Prepared by the Canadian Institute for Advanced Research, this is a fairly decent-sized dataset. The major advantage of this dataset is that it contains a balanced distribution of 10 non-overlapping categories. The images are of low resolution and size, and hence the dataset can be used on for training on fairly small memory footprint systems.
Building an image classifier
CIFAR-10 is one of the few...
