Install any recent version of Python, tested on Python 3.8.
python -m pip install --upgrade pip
python -m pip install -r requirements.txtIf you only want to greyscale the CelebA dataset, run the greyscaling218x178.py file. This will generate folders with training, validation and test data. This script will preserve the original resolution and aspect ratio.
If you want to change the resolution to 256x256 and add zero-padding to keep the aspect ratio, run the greyscaling256x256.py script. This script might be useful for certain models that require a specific input.
If you want to convert a padded image (256x256) to its original resolution (218x178) and aspect ratio, use the padded2original.py script.
If you want to calculate the FID score, L2 Norm and Cosine Similarity between two folders of images, run the evaluate.py script. This script can also be accessed via the CLI.
Usage:
python evaluate.py --path-predicted <path to predicted images> --path-ground-truth <path to ground truth images>Example:
python evaluate.py --path-predicted ./unpadded --path-ground-truth ./input/218/test/original