Generative adversarial networks for histology images implemented using PyTorch + torchsupport. Based on the histoGAN Matlab implementation.
To install histogan for usage as a python package and to install all required dependencies, please run:
python setup.py install
in your python environment.
Run the train.py script on a directory containing training data, e.g.:
python train.py --prefix histogan-runs --name histogan --batch-size 32 msi-data
to train a GAN on image tiles from the training directory msi-data at batch-size 32, saving checkpoints in histogan-runs/histogan. The size of the generated images is determined by the size of images in the training dataset.
The data directory should contain the following subdirectories:
data
|
----MSI
|
----MSS
containing tiles of H&E-stained slides of MSI tumours and MSS tumours respectively. The actual names of those subdirectories are irrelevant, just make sure that your data directory contains all MSI tiles in one folder and all MSS tiles in the other.
More generally, the train.py script can handle datasets of the form
data
|
----CLASS 1
|
----CLASS 2
|
...
|
----CLASS N