Abstract:The common occurrence of occlusion-induced incompleteness in point clouds has made point cloud completion (PCC) a highly-concerned task in the field of geometric processing. Existing PCC methods typically produce complete point clouds from partial point clouds in a coarse-to-fine paradigm, with the coarse stage generating entire shapes and the fine stage improving texture details. Though diffusion models have demonstrated effectiveness in the coarse stage, the fine stage still faces challenges in producing high-fidelity results due to the ill-posed nature of PCC. The intrinsic contextual information for texture details in partial point clouds is the key to solving the challenge. In this paper, we propose a high-fidelity PCC method that digs into both short and long-range contextual information from the partial point cloud in the fine stage. Specifically, after generating the coarse point cloud via a diffusion-based coarse generator, a mixed sampling module introduces short-range contextual information from partial point clouds into the fine stage. A surface freezing modules safeguards points from noise-free partial point clouds against disruption. As for the long-range contextual information, we design a similarity modeling module to derive similarity with rigid transformation invariance between points, conducting effective matching of geometric manifold features globally. In this way, the high-quality components present in the partial point cloud serve as valuable references for refining the coarse point cloud with high fidelity. Extensive experiments have demonstrated the superiority of the proposed method over SOTA competitors. Our code is available at <a class="link-external link-https" href="https://github.com/JS-CHU/ContextualCompletion" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the incompleteness of point clouds in 3D point cloud completion (PCC, Point Cloud Completion) caused by occlusion and other reasons. Specifically, existing PCC methods face challenges in generating fine - textured details, especially in terms of high - fidelity results. Therefore, this paper proposes a new high - fidelity 3D point cloud completion method, aiming to improve the completion effect in the fine - grained stage by mining local and global context information. ### Problem Background 3D point clouds are a widely used data structure for representing the geometric shapes of objects. However, due to factors such as occlusion, surface reflectivity, and scanning range limitations, the original point clouds usually have missing regions, which will affect the quality of 3D models and the performance of advanced tasks (such as classification, segmentation, and object detection). Therefore, point cloud completion (PCC) has become an important research topic. ### Shortcomings of Existing Methods Existing PCC methods usually adopt a coarse - to - fine paradigm, that is, first generate a rough complete point cloud and then gradually refine the texture details. Although diffusion models perform well in the coarse stage, it is still difficult to generate high - fidelity results in the fine - grained stage because PCC is essentially an ill - posed problem. In addition, existing methods fail to fully utilize the inherent context information in the partial point clouds, especially the similarity of rigid - transformation - invariant features between local regions. ### Solution This paper proposes a new high - fidelity PCC method to improve the completion effect in the fine - grained stage by mining short - range and long - range context information in the partial point clouds. Specifically, it includes the following modules: 1. **Hybrid Sampling Module**: Seamlessly fuse the roughly generated point cloud with the partial point cloud to introduce short - range context information. 2. **Surface Freezing Mechanism**: Protect the points from the noise - free partial point cloud from interference. 3. **Similarity Modeling Module**: Design a learnable rigid - transformation module to extract rigid - transformation - invariant geometric manifold features for effective global matching. ### Method Overview The entire method is divided into two stages: - **Coarse Generation Stage**: Use a diffusion - model - based coarse generator (DCG) to generate a rough but complete point cloud based on the partial point cloud. - **Fine - grained Optimization Stage**: Through a context - aware optimizer (CRef), use short - range and long - range context information to refine the rough point cloud and generate a point cloud with a complete shape and high - fidelity texture details. ### Experimental Results Experiments show that this method outperforms existing methods on multiple datasets, especially on datasets such as ShapeNet - ViPC, 3D - EPN, PartNet, and MVP. Through quantitative and qualitative comparison experiments, the effectiveness and superiority of the proposed method are verified. ### Summary This paper proposes a new high - fidelity 3D point cloud completion method. By mining short - range and long - range context information in the partial point clouds, it solves the challenge of existing methods in generating high - fidelity results in the fine - grained stage. Experimental results show that this method has achieved excellent performance on multiple datasets.

Digging into Intrinsic Contextual Information for High-fidelity 3D Point Cloud Completion

Aggregating Feature Point Cloud for Depth Completion

Dual-scale Point Cloud Completion Network Based on High-Frequency Feature Fusion

Adaptive Recurrent Forward Network for Dense Point Cloud Completion

View-Guided Point Cloud Completion

Leveraging Single-View Images for Unsupervised 3D Point Cloud Completion

Point Cloud Completion Via Self-projected View Augmentation and Implicit Field Constraint

P2C: Self-Supervised Point Cloud Completion from Single Partial Clouds

N-DPC: Dense 3D Point Cloud Completion Based on Improved Multi-Stage Network

Multi-Dimensional Graph Interactional Network for Progressive Point Cloud Completion

Incomplete Laser 3D Point Cloud Classification and Completion Approach Based on Object Symmetry under Occlusion Conditions

PointCFormer: a Relation-based Progressive Feature Extraction Network for Point Cloud Completion

Position-aware Guided Point Cloud Completion with CLIP Model

CRA-PCN: Point Cloud Completion with Intra- and Inter-level Cross-Resolution Transformers

PCUNet: A Context-Aware Deep Network for Coarse-to-Fine Point Cloud Completion

PCDreamer: Point Cloud Completion Through Multi-view Diffusion Priors

CONTEXTUAL CLASSIFICATION OF POINT CLOUD DATA BY EXPLOITING INDIVIDUAL 3D NEIGBOURHOODS

High-Fidelity Point Cloud Completion with Low-Resolution Recovery and Noise-Aware Upsampling

Multi-Scope Feature Extraction for Point Cloud Completion

3DPCTN: Two 3D Local-Object Point-Cloud-Completion Transformer Networks Based on Self-Attention and Multi-Resolution

A Survey of Point Cloud Completion