DeepLearning-StainTransfer

Replication and extension of GAN-based virtual H&E staining of skin tissues. Includes model training, evaluation, and exploratory analysis.

Running KID-FID-SSIM

To run the KID, FID, SSIM metrics, make sure you have installed the following packages:

numpy
torch
torchmetrics
torchvision

Also, you should arrange the directories of test directories, and dataset itself including the unstained and H&E stained images.

directories = {"DCLGAN":"DCL_TEST_DIR", 
               "CycleGAN": "Cycle_TEST_DIR",
               "CutGAN": "CUT_TEST_DIR"}

generated_image_path =  directories[model]
stained_image_path = '../data/stained/'
unstained_image_path = '../data/unstained/'

After putting the correct directories, you could run by python fid_kid_ssim.py --variation stained-vstained --model DCLGAN --metric fid. You could see the other options with python fid_kid_ssim.py --help.

Weights of the Trained Models

CycleGAN

The model needed to generate stained images is available under the checkpoints folder.

DCLGAN

You can download the model's weights via this link. It is located in the 'run_50' folder.

CUTGAN

...