This is the code and file for training zlineSegementation

readimagefiles reads images in overlay folder and mask folder
overlay images were generated with 1 cardiomyocyte image overlaying with 17 different fibroblast images, therefore, every 17 overlay images
correspond to 1 mask image

cross_valid uses k-fold validation method for training, it takes sample and mask data, fold, and model. It allows continuing training of 
previously paused training as long as indicate the fold number.

k_cross_val_split first shuffle data, and put data into k bins.

model_train contains training for model, whether continue previously pause training or start fresh. it stores the best model with lowest 
validation error, and record all losses for later plotting if necessary

loss contains dice loss, which can be adjusted with weight depending on whether heavy penalize on false positive/negative is necessary