In this chapter, we will fine-tune a VGG model that has been trained on ImageNet data with 1,000 classes. We have provided an initial project with a pretrained VGG model and some utility files. You can go ahead and download the code from https://github.com/mlwithtf/mlwithtf/tree/master/chapter_09.

In the folder chapter-09, you will have the following structure:

- data
--VGG16.npz
- samples_data
- production
- utils
--__init__.py
--debug_print.py
- README.md

There are two files that you should understand:

• VGG16.npz is the pre-trained model that is exported from the Caffe model. Chapter 11, Going Further - 21 Problems will show you how to create this file from the Caffe model. In this chapter, we will use this as the initial values for our model. You can download this file from the README.md in the chapter_09 folder.
• production is the Flask server that we created to serve as a web interface for users to upload and correct the model.
• debug_print.py contains some methods that we will use during this chapter to understand the network structure.
• samples_data contains some images of cats, dogs, and cars that we will use throughout the chapter.