PARIS: Part-level Reconstruction and Motion Analysis for Articulated Objects

TL;DR

PARIS is a self-supervised, end-to-end method for part-level reconstruction and motion analysis of articulated objects from multi-view images, outperforming prior methods without requiring 3D supervision.

Contribution

It introduces a novel architecture that jointly learns shape, appearance, and motion parameters for articulated objects without supervision.

Findings

Improves shape reconstruction with a Chamfer-L1 distance reduction of 3.94 (45.2%) for objects.

Achieves 26.79 (84.5%) reduction in part reconstruction error.

Attains 5% error rate in motion estimation across 10 categories.

Abstract

We address the task of simultaneous part-level reconstruction and motion parameter estimation for articulated objects. Given two sets of multi-view images of an object in two static articulation states, we decouple the movable part from the static part and reconstruct shape and appearance while predicting the motion parameters. To tackle this problem, we present PARIS: a self-supervised, end-to-end architecture that learns part-level implicit shape and appearance models and optimizes motion parameters jointly without any 3D supervision, motion, or semantic annotation. Our experiments show that our method generalizes better across object categories, and outperforms baselines and prior work that are given 3D point clouds as input. Our approach improves reconstruction relative to state-of-the-art baselines with a Chamfer-L1 distance reduction of 3.94 (45.2%) for objects and 26.79 (84.5%) for parts, and achieves 5% error rate for motion estimation across 10 object categories. Video summary at: https://youtu.be/tDSrROPCgUc