SuperDec: 3D Scene Decomposition with Superquadric Primitives

2025-04-01

AI Summary

Existing 3D scene representations struggle to simultaneously achieve high geometric fidelity and strong compression efficiency. Method: This paper proposes the first end-to-end learnable, scene-wide, instance-level superquadric decomposition framework. We design a local object parameter regression network, integrate it with an instance-segmentation-guided global decomposition paradigm, and employ a hybrid training strategy combining ShapeNet supervision with cross-domain generalization on ScanNet++ and Replica. Contribution/Results: Our approach establishes the first differentiable, instance-aligned mapping from raw point clouds to superquadric parameters. It significantly improves reconstruction accuracy and cross-dataset generalization. Quantitatively, it achieves state-of-the-art geometric reconstruction quality on both ScanNet++ and Replica. The learned compact superquadric representations effectively support downstream applications including robotic grasp planning and controllable 3D editing.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Learning with Manifolds
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

We present SuperDec, an approach for creating compact 3D scene representations via decomposition into superquadric primitives. While most recent works leverage geometric primitives to obtain photorealistic 3D scene representations, we propose to leverage them to obtain a compact yet expressive representation. We propose to solve the problem locally on individual objects and leverage the capabilities of instance segmentation methods to scale our solution to full 3D scenes. In doing that, we design a new architecture which efficiently decompose point clouds of arbitrary objects in a compact set of superquadrics. We train our architecture on ShapeNet and we prove its generalization capabilities on object instances extracted from the ScanNet++ dataset as well as on full Replica scenes. Finally, we show how a compact representation based on superquadrics can be useful for a diverse range of downstream applications, including robotic tasks and controllable visual content generation and editing.

Problem

• Decomposing 3D scenes into superquadric primitives
• Creating compact and expressive 3D representations
• Enabling applications in robotics and content generation

Innovation

• Decompose 3D scenes into superquadric primitives
• Use instance segmentation for scalable solutions
• Compact superquadric representation for diverse applications