CycleGAN

You can read the project explaination here: https://docs.google.com/document/d/1Eeq1lYp01QimSYfL7lo4ydQUKEBfPuumDcpRwTnM77E/edit?usp=sharing

Deployment and run

Dependencies

Run

conda env create -f environment.yml

to install the needed dependencies.

Then, activate the environment: conda activate gan

After that, install additional dependencies using: pip install flickrapi IPython opencv-python tensorflow-addons split-folders

Checkpoint

Create the directory first using mkdir -p res/checkpoints/train/.

Download the weights (link in relation) and place them in the following folder: res/checkpoints/train/.

Datasets

Create the directory first using mkdir -p res/dataset/x/ and mkdir -p res/dataset/y/.

Download the datasets (link in relation) and place them in order to have the following structure (create the train/test/val directories first, like before):

portrait task: res/dataset/x/train/image0.jpg and res/dataset/y/train/image0.jpg
flower task: res/dataset/x/train/jpg/image0.jpg and res/dataset/y/train/jpg/image0.jpg

Note that the flower task needs a jpg/ additional folder.

You can manually divide the datasets into train/test/val or use the script in utils/splitter.py

Run

Run the notebook.ipynb using jupyter notebook command.

Other scripts

The documentation of other (less important) scripts is a work in progress.