Creating a simple image regressor with AutoKeras

The power and usefulness of AutoKeras is not limited to image classification. Although not as popular, image regression is a similar problem where we want to predict a continuous quantity based on the spatial information in an image.

In this recipe, we'll train an image regressor to predict people's ages while using AutoML.

Let's begin.

Getting ready

We'll be using APPA-REAL dataset in this recipe, which contains 7,591 images labeled with the real and apparent ages for a wide range of subjects. You can read more about the dataset and download it from http://chalearnlap.cvc.uab.es/dataset/26/description/#. Decompress the data in a directory of your preference. For the purposes of this recipe, we'll assume the dataset is located within the ~/.keras/datasets/appa-real-release folder.

Here are some sample images:

Figure 11.1 – Sample images from the APPA-REAL dataset

...