The implemented architecture is taken from the work in . It performs an end-to-end segmentation. The number of classes used in this specific model is 2. It can be modified by changing the number of filters in the last convolutional layer based on your needs.
.Tran Minh Quan, David G. C. Hildebrand, and Won-Ki Jeong. FusionNet: A Deep Fully Residual Convolutional Neural Network for Image Segmentation in Connectomics. CoRR, abs/1612.05360, 2016.