The main concept behind the inception network is to combine different convolutions in a single layer. The combination is done by combining 7x7, 5x5, 3x3, and 1x1 convolutions to give to the next layer. Through this, the network can extract more features of the network and thus give better accuracy. This is shown in the following image of the Google inception V3 network. You can try to access the code at chapter_08/nets/inception_v3.py.

The image is taken from https://github.com/tensorflow/models/blob/master/research/inception/g3doc/inception_v3_architecture.png