reduced-8 uses the same training data as semantic-8 but only a small subset of test data, so that computationally demanding algorithms can also compete. Class labels are again {1: man-made terrain, 2: natural terrain, 3: high vegetation, 4: low vegetation, 5: buildings, 6: hard scape, 7: scanning artefacts, 8: cars}. An additional label {0: unlabeled points} marks points without ground truth and should not be used for training! Uniform downsampling with a resolution of 0.01 m was performed on the test set. Manipulations of the training set are left to the users. Please also try to participate in the semantic-8 benchmark!

We provide both training and test data as zipped ascii text files with format {x, y, z, intensity, r, g, b}. The ground truth is provided as single column ascii file, where the row ids of the class labeles and the points correspond. 7zip was used for compression and is available for both windows and linux.