Abstract:Point clouds are commonly used in various practical applications such as autonomous driving and the manufacturing industry. However, these point clouds often suffer from incompleteness due to limited perspectives, scanner resolution and occlusion. Therefore the prediction of missing parts performs a crucial task. In this paper, we propose a novel method for point cloud completion. We utilize a spherical template to guide the generation of the coarse complete template and generate the dynamic query tokens through a correspondence pooling (Corres-Pooling) query generator. Specifically, we first generate the coarse complete template by embedding a Gaussian spherical template into the partial input and transforming the template to best match the input. Then we use the Corres-Pooling query generator to refine the coarse template and generate dynamic query tokens which could be used to predict the complete point proxies. Finally, we generate the complete point cloud with a FoldingNet following the coarse-to-fine paradigm, according to the fine template and the predicted point proxies. Experimental results demonstrate that our T-CorresNet outperforms the state-of-the-art methods on several benchmarks. Our Codes are available at <a class="link-external link-https" href="https://github.com/df-boy/T-CorresNet" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### Problems Addressed by the Paper The paper aims to address the issue of incompleteness in point cloud data in practical applications (such as autonomous driving and manufacturing) due to limited viewpoints, insufficient scanner resolution, and occlusion. Specifically, the authors propose a novel point cloud completion method—T-CorresNet, which uses spherical templates and a corresponding pooling query generation strategy to predict the missing parts, thereby generating a complete point cloud. ### Main Contributions 1. **Proposed a novel T-CorresNet**: Utilizes spherical templates to guide the generation of a rough complete point cloud, enabling the network to better understand the global shape of the model. 2. **Designed a corresponding pooling (Corres-Pooling) query generator**: Generates dynamic query tokens, retains the original information of the input point cloud, and enhances the capability of the predicted point proxies. 3. **Experimental results**: T-CorresNet outperforms existing state-of-the-art methods on multiple commonly used benchmarks. ### Method Overview 1. **Spherical Template Guided Encoder**: - Expands the input space using a standard Gaussian spherical template to generate a rough complete template. - Converts the input point cloud into abstract feature vectors (point proxies) through a serialized token generator and embeds positional information. - Models the feature relationships between points using graph transformer encoder blocks to generate a rough complete template. 2. **Corresponding Pooling Query Generation**: - Identifies points highly similar to the input through a corresponding attention module and replaces these points with parts of the input to form a candidate point pool (Corres-Pool). - Generates fine templates and dynamic query tokens for predicting the point proxies of the complete point cloud. 3. **Dynamic Transformer Decoder**: - Models the feature relationships between points using kNN, self-attention, and cross-attention mechanisms. - Generates value vectors suitable for query tokens and uses FoldingNet to recover detailed local shapes of the fine template. 4. **Loss Function**: - Uses Chamfer distance as the loss function to measure the distance between two unordered point sets. ### Experimental Results - Quantitative and qualitative experimental results on PCN, ShapeNet-55, and ShapeNet-34 benchmarks show that T-CorresNet outperforms existing state-of-the-art methods in most categories. - Visualization results indicate that T-CorresNet performs better in detail retention, especially when handling complex shapes. ### Conclusion T-CorresNet effectively addresses the incompleteness issue in point cloud completion tasks by introducing spherical templates and a corresponding pooling query generation strategy, demonstrating potential in practical applications.

T-CorresNet: Template Guided 3D Point Cloud Completion with Correspondence Pooling Query Generation Strategy

Adaptive Recurrent Forward Network for Dense Point Cloud Completion

RFNet - Recurrent Forward Network for Dense Point Cloud Completion.

Point Cloud Completion Cascade Optimization Network Based on Feature Fusion

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

SD-Net: Spatially-Disentangled Point Cloud Completion Network

DF-Net: Dynamic and Folding Network for 3D Point Cloud Completion

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

SoftPoolNet: Shape Descriptor for Point Cloud Completion and Classification

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

Part-Aware Point Cloud Completion Through Multi-Modal Part Segmentation.

ProxyFormer: Proxy Alignment Assisted Point Cloud Completion with Missing Part Sensitive Transformer

A novel method for point cloud completion: Adaptive region shape fusion network

Completing Partial Point Clouds with Outliers by Collaborative Completion and Segmentation

PDP-NET: Patch-Based Dual-Path Network for Point Cloud Completion

Sequence Generation Completion Method and Resolution Scaling Network for Point Cloud Completion

GRNet: Gridding Residual Network for Dense Point Cloud Completion

CarvingNet: Point Cloud Completion by Stepwise Refining Multi-Resolution Features

ProtoComp: Diverse Point Cloud Completion with Controllable Prototype

Point Cloud Completion by Dynamic Transformer with Adaptive Neighbourhood Feature Fusion

A Cascaded Graph Convolutional Network for Point Cloud Completion