Abstract:The acquisition of point clouds is usually accompanied by noise due to imperfect laser scanning or image-based reconstruction techniques. Deep learning-based methods have achieved impressive performance in point cloud denoising. However, the features captured by a denoising network from noisy point clouds are usually contaminated by noise during training. The feature noise will lead to the oscillation of back-propagated gradients, which interferes with parameter optimization and reduces the denoising performance. In this paper, we propose to explicitly clean up feature noise for point cloud denoising from two aspects: feature noise cleaning and network training. From the first aspect, we propose the feature clean network (FCNet for short) to explicitly clean up the feature noise. From the second aspect, we train FCNet by a teacher-student learning model to learn the noise-free features under the guidance of feature domain losses. Specifically, FCNet is designed with emphasis on two modules: non-local self-similarity (NSS) and weighted average pooling (WAP). NSS module smooths features through a non-local filter based on the inherent non-local self-similarity of point clouds. WAP module applies original weights calculated by the statistical outlier removal algorithm to suppress the feature noise induced by outliers. In the teacher-student learning model, we introduce a clean input using the noisy point and its clean neighbors. The teacher network accepts the clean input to capture noise-free features. The student network is trained to imitate the teacher network to learn noise-free features by minimizing the feature loss. The experiments on synthetic and real scanned point clouds show that FCNet outperforms state-of-the-art point cloud denoising methods.

RePCD-Net: Feature-Aware Recurrent Point Cloud Denoising Network

MSPoint: Point Cloud Denoising Network Based on Multiscale Distribution Score

Dynamic Residual Dense Network For Image Denoising

RFNet - Recurrent Forward Network for Dense Point Cloud Completion.

Frequency-Relevant Residual Learning for Multi-Modal Image Denoising.

PathNet: Path-Selective Point Cloud Denoising

MODNet: Multi-offset Point Cloud Denoising Network Customized for Multi-scale Patches

Non-Local Part-Aware Point Cloud Denoising

PCDNF: Revisiting Learning-based Point Cloud Denoising via Joint Normal Filtering

Dilated Residual Encode-Decode Networks for Image Denoising

Nonlocal Low-Rank Point Cloud Denoising for 3-D Measurement Surfaces

3D Point Cloud Denoising via Deep Neural Network based Local Surface Estimation

Iterative Displacement for Point Cloud Denoising

FCNet: Learning Noise-Free Features for Point Cloud Denoising

Transformer-based Point Cloud Denoising Network

Progressive Point Cloud Denoising with Cross-Stage Cross-Coder Adaptive Edge Graph Convolution Network

3D point cloud denoising method based on global feature guidance

GeoGCN: Geometric Dual-domain Graph Convolution Network for Point Cloud Denoising

NERNet: Noise Estimation and Removal Network for Image Denoising