The dataset includes:

6788 training images and the corresponding segmentation ground truths.

5000 testing images.