Saved in:
Bibliographic Details
Main Authors: Zhang, Zeping, Laganière, Robert
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.26759
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917364684029952
author Zhang, Zeping
Laganière, Robert
author_facet Zhang, Zeping
Laganière, Robert
contents LiDAR perception is severely limited by the distance-dependent sparsity of distant objects. While diffusion models can recover dense geometry, they suffer from prohibitive latency and physical hallucinations manifesting as ghost points. We propose Scanline-Consistent Range-Aware Diffusion, a framework that treats densification as probabilistic refinement rather than generation. By leveraging Partial Diffusion (SDEdit) on a coarse prior, we achieve high-fidelity results in just 156ms. Our novel Ray-Consistency loss and Negative Ray Augmentation enforce sensor physics to suppress artifacts. Our method achieves state-of-the-art results on KITTI-360 and nuScenes, directly boosting off-the-shelf 3D detectors without retraining. Code will be made available.
format Preprint
id arxiv_https___arxiv_org_abs_2603_26759
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Physics-Aware Diffusion for LiDAR Point Cloud Densification
Zhang, Zeping
Laganière, Robert
Computer Vision and Pattern Recognition
LiDAR perception is severely limited by the distance-dependent sparsity of distant objects. While diffusion models can recover dense geometry, they suffer from prohibitive latency and physical hallucinations manifesting as ghost points. We propose Scanline-Consistent Range-Aware Diffusion, a framework that treats densification as probabilistic refinement rather than generation. By leveraging Partial Diffusion (SDEdit) on a coarse prior, we achieve high-fidelity results in just 156ms. Our novel Ray-Consistency loss and Negative Ray Augmentation enforce sensor physics to suppress artifacts. Our method achieves state-of-the-art results on KITTI-360 and nuScenes, directly boosting off-the-shelf 3D detectors without retraining. Code will be made available.
title Physics-Aware Diffusion for LiDAR Point Cloud Densification
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2603.26759