A Conditional Denoising Diffusion Probabilistic Model for Point Cloud Upsampling

Wentao Qu,Yuantian Shao,Lingwu Meng,Xiaoshui Huang,Liang Xiao

2023-12-03

Abstract:Point cloud upsampling (PCU) enriches the representation of raw point clouds, significantly improving the performance in downstream tasks such as classification and reconstruction. Most of the existing point cloud upsampling methods focus on sparse point cloud feature extraction and upsampling module design. In a different way, we dive deeper into directly modelling the gradient of data distribution from dense point clouds. In this paper, we proposed a conditional denoising diffusion probability model (DDPM) for point cloud upsampling, called PUDM. Specifically, PUDM treats the sparse point cloud as a condition, and iteratively learns the transformation relationship between the dense point cloud and the noise. Simultaneously, PUDM aligns with a dual mapping paradigm to further improve the discernment of point features. In this context, PUDM enables learning complex geometry details in the ground truth through the dominant features, while avoiding an additional upsampling module design. Furthermore, to generate high-quality arbitrary-scale point clouds during inference, PUDM exploits the prior knowledge of the scale between sparse point clouds and dense point clouds during training by parameterizing a rate factor. Moreover, PUDM exhibits strong noise robustness in experimental results. In the quantitative and qualitative evaluations on PU1K and PUGAN, PUDM significantly outperformed existing methods in terms of Chamfer Distance (CD) and Hausdorff Distance (HD), achieving state of the art (SOTA) performance.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

This paper attempts to address several key issues in Point Cloud Upsampling (PCU):

1. **Generating high-quality point clouds**: Existing methods perform poorly when dealing with point clouds with blurred geometric details, such as complex structures like the eyes of animals. The proposed method can generate point clouds that are closer to the true values and have finer geometric details.

2. **Avoiding additional upsampling module design**: Traditional methods usually require the design of a dedicated upsampling module to generate dense point clouds, which increases the complexity of algorithm design. The proposed method avoids the need for an additional upsampling module by directly modeling the gradient of the data distribution.

3. **Improving robustness to noise**: Existing methods rely on Chamfer Distance (CD) loss during training, making them very sensitive to noise. The proposed method significantly improves robustness to noise through the denoising architecture of the probabilistic model.

4. **Supporting arbitrary scale point cloud generation**: Existing methods perform poorly when handling point clouds of different scales, especially in generating high-quality arbitrary scale point clouds during the inference stage. The proposed method introduces a scaling factor, enabling the generation of high-fidelity arbitrary scale point clouds during the inference stage.

Specifically, the paper proposes a Point Cloud Upsampling Network (PUDM) based on the Conditional Denoising Diffusion Probabilistic Model (C-DDPM). This network directly models the data distribution gradient of dense point clouds, fits the true values using dominant features, and achieves efficient and high-quality point cloud upsampling through the design of a dual mapping paradigm and scaling factor. Experimental results show that PUDM outperforms existing methods on multiple public datasets, particularly achieving significant improvements in metrics such as Chamfer Distance (CD) and Hausdorff Distance (HD).