Achieves high-fidelity LiDAR densification in just 156ms while strictly enforcing sensor physics to prevent 'ghost points'.
March 31, 2026
Original Paper
Physics-Aware Diffusion for LiDAR Point Cloud Densification
arXiv · 2603.26759
The Takeaway
Standard diffusion models for LiDAR are too slow for real-time autonomous driving and suffer from physical hallucinations. By treating densification as probabilistic refinement with a ray-consistency loss, this method boosts off-the-shelf 3D detectors without needing retraining.
From the abstract
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 R