Hi, this is the first version of the code.

Prerquisite packages: In order to run the program (main.py) in Python, there are some packages (PyTorch, NumPy etc.) are required and they are in the environment.yml file. (setting up environment using Anacond)

Assumption (Hypothesis): We claim that a future cross section of the retina can be interpolated from past measurements of the same cross sections. Each measurmenent, except for the last one Here, I set the number of previous measurements to 3 ("opt.num_source" argument in the options.py), we can play it.

Dataset: Data files should be under the "image_dir" argument that is set in "options.py". The image files of patients AGPS023, AGPS024, AGPS025, AGPS027, and AGPS028 are corrupted (0 KB images). You can change the train/val fractions in the fiest lines of the main.py. I initially set to 0.8/0.2. Program automatically lists the patients and sparetes the 80% of the patients for training etc. And, patients who have less than "num_source" measurements are discarded, not included to datasets. "num_source" is the number of the past measurments used to predict to future measurements. Dataset is thought as (source, target) pairs, where source is the concatenated num_source images, while target is a single image. (Each image has 3 channels, RGB).

GAN Architecture: Generator applies 2D Conv on the source image in a U-Net type, which has 2 maing sub-networks: downsampling and upsampling. And, skip connections betweek these two part copies features from the front part of the network to the rear part. Discriminator is copied from a well-known project called pix2pix. We can change its number of layers in accordance w/ the training results. We can think different architectures in future.

Code creates a project folder w/ the name of the current date, hour etc. Each project folder has its own configuration file (storing the implementation details), images folder (showing image samples), log_file/losses_over_epochs files (showing/displaying the progress of the losses throughout the training).