LiDPM: Rethinking Point Diffusion for Lidar Scene Completion

2025-04-24

AI Summary

To address the challenge of generating fine-grained geometric structures from white noise under large field-of-view conditions in outdoor large-scale LiDAR point cloud scene completion, this paper demonstrates—for the first time—that standard Denoising Diffusion Probabilistic Models (DDPMs) can be directly applied at full-scene scale without resorting to local diffusion approximations. The core method introduces a geometry-aware initial point cloud initialization strategy, integrated with a lightweight point cloud encoder and an iterative denoising decoder, enabling end-to-end scene-level generation. Evaluated on SemanticKITTI, our approach achieves state-of-the-art performance: it significantly improves completion completeness and geometric consistency, particularly enhancing detail recovery in distant and sparse regions. This work bridges the gap between object-level and scene-level point cloud diffusion modeling, establishing a novel paradigm for large-scale 3D scene generation.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Large language models for search
• Security and Privacy: Large-scale security measurements

Abstract

Training diffusion models that work directly on lidar points at the scale of outdoor scenes is challenging due to the difficulty of generating fine-grained details from white noise over a broad field of view. The latest works addressing scene completion with diffusion models tackle this problem by reformulating the original DDPM as a local diffusion process. It contrasts with the common practice of operating at the level of objects, where vanilla DDPMs are currently used. In this work, we close the gap between these two lines of work. We identify approximations in the local diffusion formulation, show that they are not required to operate at the scene level, and that a vanilla DDPM with a well-chosen starting point is enough for completion. Finally, we demonstrate that our method, LiDPM, leads to better results in scene completion on SemanticKITTI. The project page is https://astra-vision.github.io/LiDPM .

Problem

• Challenges in training lidar point diffusion models for outdoor scenes
• Bridging gap between local and object-level diffusion approaches
• Improving scene completion accuracy on SemanticKITTI dataset

Innovation

• Uses vanilla DDPM for scene completion
• Optimizes starting point for better results
• Achieves fine-grained lidar detail generation