Phil Shilane will present his preFPO on Thursday February 8 at 10AM in Room 402. The members of his committee are; Thomas Funkhouser, advisor; Adam Finkelstein and Szymon Rusinkiewicz, readers; Andrea LaPaugh and Kai Li, nonreaders. Everyone is invited to attend his talk. His abstract follows below. ---------------------------------------- Title: Shape Distinction for 3D Model Retrieval Abstract: In recent years, there has been enormous growth in the number of 3D models and their availability to a wide segment of the population. Examples include the National Design Repository which stores 3D computer-aided design (CAD) models for tens of thousands of mechanical parts, the Protein Data Bank that has atomic positions for tens of thousands of protein molecules, and the Princeton Shape Benchmark with 36,000 everyday objects represented as polygonal surface models. Given the availability of 3D data, a key problem is searching through models efficiently. Finding similar objects in a database can provide information for protein analysis or to identify objects in a robotic-vision system. As another example, by searching a database of models, even a novice user could easily create new CAD objects or graphics models by assembling them from examples, instead of the complicated task of creating models from scratch. While previous work has approached the 3D retrieval problem by creating global feature vectors for each object or by creating numerous local feature vectors, my work focuses on "distinctive" regions of a surface that distinguish a shape from objects of a different class. Distinctive regions are analyzed based on performing a shape-based search using each surface region as a query into a database. Distinctive regions of a surface have shape consistent with objects of the same type and different from objects of other types. For a classified database of models, distinction can be analyzed in a pre-processing phase, but I also show that a likelihood model can predict distinction for new shapes based on a training set. Using a priority-driven search technique that focuses on the most distinctive regions of each shape provides better search results than any previously published work. While the current definition of distinction is based on shape retrieval, it is useful for other graphics applications such as mesh visualization, icon generation, and mesh simplification. I also propose a new methodology to analyze shape retrieval by creating a common dataset of classified 3D models and software tools called the Princeton Shape Benchmark (PSB). The freely available PSB includes the models, classification, annotations for each model, and a suite of software tools to evaluate and analyze shape-retrieval results. My experiments with these classifications expose different properties of shape-based retrieval methods that were not previously known. As an example, shape descriptors that had poor performance when considering all models in the PSB actually had the best performance on certain classes of models. Perhaps the largest contribution of the PSB is that the dataset and software tools make it easier to perform shape-similarity research. Undergraduate computer graphics courses, senior thesis work, and doctoral projects have experimented with the 3D models and tools making the PSB a de-facto standard for measuring shape-similarity methods in the Computer Graphics community.