The Princeton Shape Benchmark

In recent years, many shape representations and geometricalgorithms have been proposed for matching 3D shapes.Usually, each algorithm is tested on a different (small)database of 3D models, and thus no direct comparison isavailable for competing methods.In this paper, we describe the Princeton Shape Benchmark(PSB), a publicly available database of polygonalmodels collected from the World Wide Web and a suite oftools for comparing shape matching and classification algorithms.One feature of the benchmark is that it providesmultiple semantic labels for each 3D model. For instance, itincludes one classification of the 3D models based on function,another that considers function and form, and othersbased on how the object was constructed (e.g., man-madeversus natural objects).We find that experiments with these classifications canexpose different properties of shape-based retrieval algorithms.For example, out of 12 shape descriptors tested,Extended Gaussian Images [13] performed best for distinguishingman-made from natural objects, while they performedamong the worst for distinguishing specific objecttypes. Based on experiments with several different shapedescriptors, we conclude that no single descriptor is bestfor all classifications, and thus the main contribution of thispaper is to provide a framework to determine the conditionsunder which each descriptor performs best.