Artificial Intelligence for Robotics

Processing the image

Now we have several steps to take before we can present our data to the neural network for training. Our camera on the robot has way too much resolution for what we need for the network, and we want to use the minimum amount of data in the neural network we can get away with:

Our first step is to downsample the image to 320 x 200. We cut the image in half, and keep only the top half, which eliminates the floor from our consideration. We resize the rectangular image to 244 x 244, which is an appropriate size for our neural network to process. We convert the image to black and white, so that we only have one channel to ...