Data originally from http://mi.eng.cam.ac.uk/research/projects/VideoRec/CamVid/
Training-validation-test splits from https://github.com/alexgkendall/SegNet-Tutorial
There are 31+1 semantic classes (+1 comes from the Void class). They are usually grouped into 11+1 classes: Sky, Buliding, Pole, Road, Sidewalk, Tree, SignSymbol, Fence, Vehicle, Pedestrian, Bicyclist, Void. A python class for loading the dataset with 11 classes can be found here (search for CamVid): https://github.com/Ivan1248/Vidlu/blob/master/vidlu/data/datasets/datasets.py