IPCD: Intrinsic Point-Cloud Decomposition

2025-11-13

AI Summary

Point clouds’ unstructured geometry and absence of explicit global illumination modeling hinder accurate separation of albedo and shadows, limiting photorealistic applications such as relighting and texture editing. To address this, we propose IPCD-Net—the first intrinsic decomposition network for point clouds that explicitly models global illumination direction. Leveraging a Projection-based Luminance Distribution (PLD) prior, IPCD-Net achieves high-fidelity decomposition on unstructured inputs via point-wise feature aggregation, multi-view projection, and hierarchical optimization. Trained on synthetic data and validated on real-world scenes, our method significantly suppresses cast shadows in estimated albedo, improves shadow color fidelity, and enhances cross-illumination registration accuracy. This work establishes a generalizable foundation for point cloud relighting and editing in AR and robotic applications.

Technology Category

• Computer Vision: Computational Photography, Image & Video Synthesis
• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Calibration & Uncertainty Quantification

Application Category

• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for incomplete, noisy, or partially observed Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

Point clouds are widely used in various fields, including augmented reality (AR) and robotics, where relighting and texture editing are crucial for realistic visualization. Achieving these tasks requires accurately separating albedo from shade. However, performing this separation on point clouds presents two key challenges: (1) the non-grid structure of point clouds makes conventional image-based decomposition models ineffective, and (2) point-cloud models designed for other tasks do not explicitly consider global-light direction, resulting in inaccurate shade. In this paper, we introduce extbf{Intrinsic Point-Cloud Decomposition (IPCD)}, which extends image decomposition to the direct decomposition of colored point clouds into albedo and shade. To overcome challenge (1), we propose extbf{IPCD-Net} that extends image-based model with point-wise feature aggregation for non-grid data processing. For challenge (2), we introduce extbf{Projection-based Luminance Distribution (PLD)} with a hierarchical feature refinement, capturing global-light ques via multi-view projection. For comprehensive evaluation, we create a synthetic outdoor-scene dataset. Experimental results demonstrate that IPCD-Net reduces cast shadows in albedo and enhances color accuracy in shade. Furthermore, we showcase its applications in texture editing, relighting, and point-cloud registration under varying illumination. Finally, we verify the real-world applicability of IPCD-Net.

Problem

• Separating albedo from shade in colored point clouds
• Handling non-grid structure of point clouds effectively
• Capturing global-light direction for accurate decomposition

Innovation

• IPCD-Net extends image models for point-cloud decomposition
• Projection-based Luminance Distribution captures global light direction
• Hierarchical feature refinement processes multi-view projection data