notes.txt

Steps for training
==================

1. Gather the training set and put it into the proper directory structure described in
   temp_combine_datasets.sh (A/train/images B/train/images)
   - A's images will be the generated images
   - B's images will be the cleaned up, "true" images hand-crafted by an RSE intern

2. Run the temp_combine_datasets.sh script to combine each pair of images into one pair that
   will get placed in the training folder

3. Run the training script to generate the GAN-produced images in the results folder. These
   images are the output of 1st pass of the GAN and will be used as the input for the 2nd GAN pass

4. Run the convert_1stpass_to_2ndpass_training.py script to convert the 1st pass images into
   the input images for the 2nd pass

5. Run the temp_train.sh again (this time to produce the 2nd pass images)

6. The final GAN-generated images will be in the results folder



Steps for testing
=================

1. Pretty much the same thing as training except you run the temp_test.sh script.
   Note that the testing image input has to be a concatenated image again (compose of the
   input test image concatenated with the real test image). This is because this is format
   expected by the GAN. Even though you're providing the real image to the GAN in the test
   data, don't worry. It doesn't use the real image when performing inference. It's just
   there for comparison later in the final output. Instead of concatenating the real image,
   you can concatenate any image (already tested this)