-
BSDS500 dataset
The dataset used in this paper is the BSDS500 dataset, which contains 200 natural images with over 1000 ground truth labellings. -
Additional test data
We hold out 3% of the data for validation and 5% for testing. For additional test data, we render a few PBRT-v3 test scenes that portray practical real-world scenarios1 though... -
OpenStreetMap dataset
The dataset used in this paper is a set of 50 pairs of geographic terms, ranked by 203 human subjects as ground truth.