Semantic protection method hides 3D point cloud classes while keeping usefulness
Semantic Privacy Protection with Utility Preservation for 3D Point Clouds
Computer Vision and Pattern Recognition
Summary
3D point clouds, which are data points representing objects in space, often risk exposing private information during sharing. The authors propose a way to hide the original category of these 3D shapes while still allowing others to use them for tasks like recognition and segmentation. Their method changes the data so it looks like a different category but can be reversed by authorized users. Tests show this approach reduces privacy leaks and keeps the data useful for analysis.
What this means in practice
- •For autonomous vehicle engineers: Protect original object categories in LIDAR data shared across companies while retaining usability for model training.
- •For 3d mapping providers: Share city-scale 3D scans with obfuscated original classes to protect proprietary details while enabling downstream analysis.
Authors
Jinchang zhang, Jiakai Lin, David Crandall, Guoyu Lu
Abstract
Point cloud data face serious semantic privacy risks during acquisition, transmission, and cross-institutional sharing. Existing methods mostly rely on geometric perturbation or destructive encryption, which can reduce the recognizability of the original class but often impair downstream usability. This paper proposes a class-transfer-based semantic encryption framework for point clouds, aiming to conceal original class information while preserving task utility and supporting authorized recovery. Specifically, we construct a unified latent space with a shared-backbone Normalizing Flow, and combine LoRA and FiLM to achieve parameter-efficient class-conditional adaptation. We further introduce diffusion-guided flow alignment to regularize the latent distribution, construct an energy-based category transition graph, and obtain an optimal class-transfer table through global matching. Then, a latent-space Neural ODE continuously evolves source-class latents into target-class latents, which are decoded into target-class point clouds through the inverse flow. We adopt attacker-oriented metrics, including New-Class Recognition Rate (NCRR), Original-Class Leakage Rate (OCLR), and Original Label Recovery Rate (OLRR), to evaluate privacy and utility. Experiments on classification and segmentation benchmarks show that the proposed method achieves controllable semantic transformation, effectively reduces original-class semantic leakage, preserves downstream learnability in the protected domain, and supports reliable authorized reconstruction.