Learning Convex Decomposition via Feature Fields

TL;DR

This paper introduces a novel feature learning approach for convex decomposition of 3D shapes, enabling a scalable, self-supervised, open-world model that produces high-quality decompositions across diverse representations.

Contribution

It presents the first feed-forward, self-supervised model for open-world convex decomposition using feature fields, improving quality and generalization over previous methods.

Findings

Decompositions outperform existing methods in quality

Model generalizes across different shape representations

Enables scalable learning on large datasets

Abstract

This work proposes a new formulation to the long-standing problem of convex decomposition through learning feature fields, enabling the first feed-forward model for open-world convex decomposition. Our method produces high-quality decompositions of 3D shapes into a union of convex bodies, which are essential to accelerate collision detection in physical simulation, amongst many other applications. The key insight is to adopt a feature learning approach and learn a continuous feature field that can later be clustered to yield a good convex decomposition via our self-supervised, purely-geometric objective derived from the classical definition of convexity. Our formulation can be used for single shape optimization, but more importantly, feature prediction unlocks scalable, self-supervised learning on large datasets resulting in the first learned open-world model for convex decomposition. Experiments show that our decompositions are higher-quality than alternatives and generalize across open-world objects as well as across representations to meshes, CAD models, and even Gaussian splats. https://research.nvidia.com/labs/sil/projects/learning-convex-decomp/