Training and freezing an image captioning model

In this section, we'll first summarize the process of training the Show and Tell model called im2txt, documented at https://github.com/tensorflow/models/tree/master/research/im2txt, with some tips to help you better understand the process. Then we'll show some key changes to the Python code that comes with the im2txt model project, in order to freeze the model to be ready for use on mobile devices.

Training and testing caption generation

If you have followed the Setting up the TensorFlow Object Detection API section in Chapter 3, Detecting Objects and Their Locations, then you already have the im2txt folder installed; otherwise simply cd to your TensorFlow source root...