October 2018
Intermediate to advanced
472 pages
10h 57m
English
Imagine a situation where you might want to build a convolution classifier on a small set of images. The problem here is that the classifier will easily overfit on this small set of data. The reason why the classifier will overfit is that there are very few images that are similar. That is, there are not a lot of variations for the model to capture within a specific class so that it can be robust and perform well on new data.
Keras provides a preprocessing utility called ImageDataGenerator that can be used to augment image data with simple configuration.
Its capabilities include the following:
Read now
Unlock full access