My first attempt at image recognition using a training set I assembled myself. Despite using a very simple neural network and a relatively small set of training images I'm pleasantly surprised with the 91% accuracy on the training data.
I decided to throw in some images of things not in one of the 5 training classes just for fun and to see how the model would react.
Training images downloaded with Bing Image Search API
Packages used include:
numpy
cv2
PIL
matplotlib
tensorflow