Department of Computer Engineering
CS 590 SEMINAR
Affordance Based Scale Normalization on Shape Collections
Halil İbrahim Kayım
Computer Engineering Department
With the increasing number of 3D shape collections, retrieving individual shapes from a collection of shapes becomes a challenging problem. As opposed to using the geometrical or topological properties of such shapes, a slightly newer technique is to use their affordances, i.e. the way external agents interact with them. Most of the techniques that make use of affordances have a strong assumption that the shapes in these collections to be at a certain scale and orientation. We propose a scale normalization technique using the affordance of shape collections, so that they can be indexed on and retrieved from shape retrieval systems. Our initial study consists of re-scaling chair models, whose affordances are known to be sitting. In the particular case of chair models, we plan to first find the surface on which an agent can sit, and then re-scale the models in such a way that the surface does not exceed a certain height compared to the agent. We then plan to extend this study to include different shape collections, such as sofas, beds, or tables, different affordances such as lying down or standing, and different agent percentiles.
DATE: 03 April, 2017, Monday @ 16:05