Google offers several scripts and tutorials explaining how to directly use its inception networks, or how to retrain them for new applications. The directory dedicated to this architecture in the tensorflow/models Git repository (https://github.com/tensorflow/models/tree/master/research/inception) is also rich and well-documented. Moreover, a pretrained version of Inception V3 is available on TensorFlow Hub, which gives us the opportunity to introduce this platform.
TensorFlow Hub is a repository of pretrained models. In a similar way to how Docker allows people to easily share and reuse software packages, removing the need to reconfigure distributions, TensorFlow Hub gives access to pretrained models so that people do not have to spend time and resources reimplementing and retraining. It combines a website (https://tfhub.dev) where people can search for specific models (depending, for example, on the target recognition task), and a Python package...
