Update

change to be compatible with the resnet18 in PyTorch 1.2
remove tensorboardX

Config

pytorch 1.2.0
dataset: ICIAR 2018, please download via the grand-challenge website.

To split the dataset into 10-fold, please run the following commind in the "dataset2018" folder:

python split_dataset.py

Single fold model

To train the patch-wise network:

$ python train.py --network=1 --patches-overlap=0 --fold-index=1

Config datset (2015/2018) and checkpoint path in option.py

To evaluate the patch-wise network:

$ python validate.py --network=1 --patches-overlap=0 --fold-index=1  

To train the image-wise network:

$ python train.py --network=2 --patches-overlap=0 --fold-index=1

Config datset and checkpoint path in option.py

To evaluate the image-wise netowork:

$ python validate.py --network=2 --patches-overlap=0 --fold-index=1

Cross validation

To run the 10-fold cross-validatoin (based on the trained models):

$ python corss_evaluate.py