here is my implementation of the basic googLeNet describe in the paper "Going Deeper with Convolutions" in keras - View it on GitHub