To train a TensorFlow Classifier using Docker in Ubuntu you first must have the following prerequisites before starting:
Image datasets that you will be training the classifier for (You may use Image-Net to acquire large amounts of images quickly that are of research grade to train your classifier, please note that this database doesn't have EVERYTHING and thus if you can't find something then you will need to acquire it elsewhere)
Clean your image datasets of "dirty" images
Notes:
This documentation is assuming that you have Docker installed in Ubuntu already, you may look up a quick guide on installing Docker (very simple)
Starting:
In your Ubuntu home directory create a directory called: tf_prod (this will be the production directory for TensorFlow)
Now in tf_prod you will create directories that will be themes of the images you will be classifying. e.g. "Plants", "Trees", "Cars"
Now in those directories you will create specific directories in them such as in "Plants" you will create a directory called "Hibiscus"
Now load in your images into the specific directories you have created