Self-Supervised Point Cloud Completion based on Multi-View Augmentations of Single Partial Point Cloud

2025-09-26

AI Summary

Point cloud completion faces three key challenges: supervised methods suffer from poor generalization due to synthetic-to-real domain shift; unsupervised/weakly supervised approaches rely on complete ground-truth point clouds or multi-view inputs; and existing self-supervised methods yield suboptimal reconstruction quality owing to weak supervisory signals. To address these, we propose the first self-supervised framework for single-view partial point cloud completion. Our method introduces (1) a multi-view geometric consistency augmentation strategy to generate strong self-supervision, and (2) the first integration of the state-space model Mamba into point cloud generation, enabling effective long-range dependency modeling and enhanced feature representation. Extensive experiments on synthetic (ShapeNet) and real-world (KITTI, SemanticKITTI) benchmarks demonstrate state-of-the-art performance, with significant improvements in completeness, geometric fidelity, and cross-domain generalization.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: State Estimation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Point cloud completion aims to reconstruct complete shapes from partial observations. Although current methods have achieved remarkable performance, they still have some limitations: Supervised methods heavily rely on ground truth, which limits their generalization to real-world datasets due to the synthetic-to-real domain gap. Unsupervised methods require complete point clouds to compose unpaired training data, and weakly-supervised methods need multi-view observations of the object. Existing self-supervised methods frequently produce unsatisfactory predictions due to the limited capabilities of their self-supervised signals. To overcome these challenges, we propose a novel self-supervised point cloud completion method. We design a set of novel self-supervised signals based on multi-view augmentations of the single partial point cloud. Additionally, to enhance the model's learning ability, we first incorporate Mamba into self-supervised point cloud completion task, encouraging the model to generate point clouds with better quality. Experiments on synthetic and real-world datasets demonstrate that our method achieves state-of-the-art results.

Problem

• Reconstructing complete shapes from partial point clouds
• Overcoming limitations of supervised and unsupervised completion methods
• Enhancing self-supervised completion using multi-view augmentations and Mamba

Innovation

• Multi-view augmentations generate self-supervised signals
• Mamba model enhances point cloud generation quality
• Self-supervised completion using single partial input