Hi Aleko,

There can be multiple factors for such a mask. May I know the number of images in training and testing, batch size, image size, number of classes, number of Epochs, and threshold you have used in the train.py.