1. Downloading and preparing data.

   1. Download data from here
   2. Unpack it and move it to the root of the project
   3. Rename "archive" folder into "data" folder
   4. Create "data_transformed" folder
   5. Create "mask_to_no_mask", "no_mask_to_mask" and "cycle_transform" folders inside "data_transformed" folder
   6. With pretrained Cycle GAN create normal-LFW and put it in the "mask_to_no_mask" folder
   7. With pretrained Cycle GAN create masked-LFW and put it in the "no_mask_to_mask" folder
   8. With pretrained Cycle GAN create cycle-LFW and put it in the "cycle_transform" folder Note the structure should be "data_transformed/A/images/*_fake.png" where A in {"cycle_transform", "mask_to_no_mask", "no_mask_to_mask"}

   ** normal-LFW, masked-LFW, cycle-LFW can be also downloaded here

2. Installing required modules

   pip install -r requirements.txt
   

Test checkpoint:

1. Name it last.ckpt
2. Put it into checkpoints folder
3. Run test script

   python -m src.train_test_identification
   

Train & test model:

python -m src.train_test_identification --train

1. Clone the repository
2. Run

mv INF634/* .

3. Download data (achieve.zip) from here
4. Put achieve.zip in the same directory where CycleGAN.ipynb resides
5. Run CycleGAN.ipynb to train, test and evaluate CycleGAN (once you ran the notebook, you will get normal-LFW, masked-LFW, cycle-LFW datasets in the folder pytorch-CycleGAN-and-pix2pix).