Shape2Motion: Joint Analysis of Motion Parts and Attributes from 3D Shapes

TL;DR

Shape2Motion is a novel method that jointly analyzes 3D shapes to segment motion parts and estimate their attributes from a single point cloud, advancing mobility analysis in 3D shape understanding.

Contribution

It introduces a joint analysis framework with two neural networks utilizing motion-driven features, enabling mobility analysis from a single 3D shape without prior segmentation or multiple models.

Findings

Achieves state-of-the-art results in motion part segmentation.

Accurately estimates motion attributes from single 3D shapes.

Introduces a new benchmark for mobility analysis.

Abstract

For the task of mobility analysis of 3D shapes, we propose joint analysis for simultaneous motion part segmentation and motion attribute estimation, taking a single 3D model as input. The problem is significantly different from those tackled in the existing works which assume the availability of either a pre-existing shape segmentation or multiple 3D models in different motion states. To that end, we develop Shape2Motion which takes a single 3D point cloud as input, and jointly computes a mobility-oriented segmentation and the associated motion attributes. Shape2Motion is comprised of two deep neural networks designed for mobility proposal generation and mobility optimization, respectively. The key contribution of these networks is the novel motion-driven features and losses used in both motion part segmentation and motion attribute estimation. This is based on the observation that the movement of a functional part preserves the shape structure. We evaluate Shape2Motion with a newly proposed benchmark for mobility analysis of 3D shapes. Results demonstrate that our method achieves the state-of-the-art performance both in terms of motion part segmentation and motion attribute estimation.