Unsupervised Co-segmentation of 3D Shapes via Functional Maps

TL;DR

This paper introduces an unsupervised approach for co-segmenting 3D shapes by leveraging functional maps to establish semantic part correspondence across a shape set, achieving competitive accuracy.

Contribution

It proposes a novel unsupervised method combining pre-segmentation and functional maps for consistent co-segmentation of 3D shapes.

Findings

Efficient co-segmentation on benchmark datasets.

Comparable accuracy to state-of-the-art methods.

Unsupervised approach reduces need for labeled data.

Abstract

We present an unsupervised method for co-segmentation of a set of 3D shapes from the same class with the aim of segmenting the input shapes into consistent semantic parts and establishing their correspondence across the set. Starting from meaningful pre-segmentation of all given shapes individually, we construct the correspondence between same candidate parts and obtain the labels via functional maps. And then, we use these labels to mark the input shapes and obtain results of co-segmentation. The core of our algorithm is to seek for an optimal correspondence between semantically similar parts through functional maps and mark such shape parts. Experimental results on the benchmark datasets show the efficiency of this method and comparable accuracy to the state-of-the-art algorithms.