Accumulative categorization: Online 3D shape classification for progressive collections

Abstract

We present a new interactive and online approach to classify 3D shapes progressively by integrating online learning and user intervention. Our system, accumulative categorization, allows users to collect, classify and annotate 3D shapes interactively in an online circular manner. Our system learns a classification model from the annotation continuously and incrementally, and, in turn, classifies newly collected shapes. During this classification process, the user is allowed to interactively correct the errors in the results based on the actual requirements. The classification model is refined based on user intervention to capture the personalized intent of categorization. With the increasing of the user interactions and the accumulation of the classified shapes, the system produces more accurate classification of the subsequent shapes by the refined classification model. Experimental results demonstrate the effectiveness of our approach.