Artificial Intelligence for Robotics

Build the toy/not toy detector

The next section covers the training program for our CNN that will tell toys apart from not toys, that is to say, it will tell us whether there is a toy in the field of view of the camera.

We start by building our training sets. We divide our pictures of the playroom into pictures with toys and pictures taken without toys. We put the toy pictures in a directory under our program directory called images/toys. We put the pictures without toys in a directory called images/not_toys. We need about 200 examples of each. So far so good:

Examples of images without toys in them (top image) and with toys in them (bottom ...