As mentioned previously, the example trained over 10 epochs is not particularly accurate. You will need to train it over many epochs to get better results. If you have been watching the cost and accuracy of the model, you'll find that cost will stay relatively flat as accuracy increased over the number of epochs, as shown in the following graph:

It is still useful to explore the results to see how the model is performing; we'll specifically look at cats:

As we can see, it currently appears to do much better with cats in very specific positions. Obviously, we need to find a solution to train it faster.