GeneGAN at its current stage takes in wsi (Whole Slide Images) in .svs
format,
samples the wsi for quality tissue samples and uses a simple DCGAN to
generate wsi images.
-
We need openslide to execute the patch generation part of the code.
brew install openslide
or Install OpenSlide (see https://openslide.org/download/) -
Install the dependencies in your local dev environment
source activate <local_dev_env>
pip install -r requirements.txt
python preprocess.py [-h] -s raw_slides_folder -p patches_output_folder [-t parsed_slides_folder] [-m masks_folder_if_exists]
python dcgan.py
The result of the trained DCGAN after about 42 epochs were as below
The Samples generated from the wsi slides look like
The different in the color of the training and generated images is because the during training, the images were normalized per channel from the overall distribution of the training images