Learning part-based templates from large collections of 3D shapes

Abstract

As large repositories of 3D shape collections continue to grow, understanding the data, especially encoding the inter-model similarity and their variations, is of central importance. For example, many data-driven approaches now rely on access to semantic segmentation information, accurate inter-model point-to-point correspondence, and deformation models that characterize the model collections. Existing approaches, however, are either supervised requiring manual labeling; or employ super-linear matching algorithms and thus are unsuited for analyzing large collections spanning many thousands of models. We propose an automatic algorithm that starts with an initial template model and then jointly optimizes for part segmentation, point-to-point surface correspondence, and a compact deformation model to best explain the input model collection. As output, the algorithm produces a set of probabilistic part-based templates that groups the original models into clusters of models capturing their styles and variations. We evaluate our algorithm on several standard datasets and demonstrate its scalability by analyzing much larger collections of up to thousands of shapes.

Learning hierarchical shape segmentation and labeling from online repositories

“…Moreover, these methods have not demonstrated the ability to identify fine-grained model structure, or hierarchies. One can rely solely on consistency in part geometry to extract meaningful segments without supervision Sidi et al 2011;Huang et al 2011;Hu et al 2012;Kim et al 2013;Huang et al 2014]. However, since these methods do not take any human input into account, they typically only detect coarse parts, and do not discover semantically salient regions where geometric cues fail to encapsulate the necessary discriminative information.…”

Meta-representation of shape families

“…The majority of these works focus on consistently segmenting sets of shapes using a range of strategies including spectral clustering [Sidi et al 2011], linear programming [Huang et al 2011], active learning [Wang et al 2012], subspace clustering [Hu et al 2012], multi-label optimization [Meng et al 2013], template fitting [Kim et al 2013], etc. A few others have focused on extracting a consistent hierarchy for the set [van Kaick et al 2013], or consistent part arrangements ] from an input family of shapes.…”

“…We follow this assumption both for the training set that defines the meta-representation and those shapes that are handled by the applications (which may not be part of the training set). Several unsupervised algorithms exist to automatically obtain such a labeled segmentation from an input set of shapes [Sidi et al 2011;Huang et al 2011;Hu et al 2012;Meng et al 2013;Kim et al 2013;Laga et al 2013], as well as semi-supervised algorithms [Wang et al 2012]. Note that most of these algorithms automatically segment the shapes and assign generic labels.…”

“…Humans typically rely on their prior knowledge to infer this information. Hence, with the growth of shape repositories, researchers have focused on co-analyzing sets of shapes in order to benefit from the collective information in the group [Kalogerakis et al 2010;Huang et al 2011;Sidi et al 2011;Wang et al 2012;Kim et al 2013]. However, the typical outcome of these methods, a segmentation and/or correspondence, does not really summarize properties that define the shapes as elements of a set.…”

Ergonomics-Inspired Reshaping and Exploration of Collections of Models

“…A number of approaches have been described on shape exploration. While such tools allow users to quickly browse large data collections, a primary goal is to find effective embeddings where interactive exploration is made possible [6], [7], [8], [9]. These methods normally start from object geometry alone and extract commonalities among a family of shapes using mechanisms such as geometric descriptors or functional maps.…”