Abstract:Unsupervised point cloud shape correspondence aims to establish point-wise correspondences between source and target point clouds. Existing methods obtain correspondences directly by computing point-wise feature similarity between point clouds. However, non-rigid objects possess strong deformability and unusual shapes, making it a longstanding challenge to directly establish correspondences between point clouds with unconventional shapes. To address this challenge, we propose an unsupervised Template-Assisted point cloud shape correspondence Network, termed TANet, including a template generation module and a template assistance module. The proposed TANet enjoys several merits. Firstly, the template generation module establishes a set of learnable templates with explicit structures. Secondly, we introduce a template assistance module that extensively leverages the generated templates to establish more accurate shape correspondences from multiple perspectives. Extensive experiments on four human and animal datasets demonstrate that TANet achieves favorable performance against state-of-the-art methods.

What problem does this paper attempt to address?

### What problem does this paper attempt to solve? This paper aims to solve the problem of **unsupervised point cloud shape correspondence**. Specifically, the author hopes to establish a point - by - point correspondence between the source point cloud and the target point cloud without labeled data. This task is especially important for dealing with non - rigid objects (such as humans and animals) with high deformability and unconventional shapes. #### Main challenges 1. **Correspondence problems of unconventional shapes**: - Non - rigid objects (such as humans and animals) have strong deformability and unusual shapes, and it is very difficult to directly establish the correspondence between point clouds. - In these cases, the complete structure of the point cloud may be destroyed or distorted due to postures such as limb adhesion and bending, resulting in the fact that directly calculating the similarity cannot guarantee an accurate correspondence. 2. **Limitations of existing methods**: - Existing methods mainly establish correspondence by calculating the similarity between point cloud features, but this method often has poor performance when dealing with unconventional shapes. - Spectral methods rely on the connectivity information of grid vertices, which is usually missing in the point cloud scenario. - Other point cloud methods directly process the original point cloud data, but are easily affected by noise and incomplete representation, resulting in matching errors. #### Proposed solutions To solve the above problems, the author proposes an **unsupervised Template - Assisted Network for point cloud shape correspondence (TANet)**, which includes two main modules: 1. **Template Generation Module**: - Generate a set of learning templates with explicit structures to provide structured guidance. - Map the template feature space to the coordinate space through the Space Aligner to ensure the structural integrity of the template. 2. **Template Assistance Module**: - Use the generated templates to establish more accurate shape correspondences from multiple perspectives. - Include Template Selector, Correlation Fusion and Transitive Consistency mechanisms to improve the accuracy of the correspondence. #### Experimental verification Through extensive experiments on four public human and animal datasets, the author proves that TANet is superior to the existing state - of - the - art methods in performance, especially when dealing with unconventional shapes. ### Summary The main contribution of this paper is to propose a new framework that uses learning templates to assist in establishing the correspondence between point clouds, thereby solving the challenges in the correspondence of unconventional shapes and achieving significant performance improvements on multiple benchmark datasets.

Unsupervised Template-assisted Point Cloud Shape Correspondence Network

Adaptive Recurrent Forward Network for Dense Point Cloud Completion

Hierarchical Shape-Consistent Transformer for Unsupervised Point Cloud Shape Correspondence

Learning Point Cloud Shapes with Geometric and Topological Structures.

AMCNet: Adaptive Matching Constraint for Unsupervised Point Cloud Registration.

Shape Self-Correction for Unsupervised Point Cloud Understanding

SE-ORNet: Self-Ensembling Orientation-aware Network for Unsupervised Point Cloud Shape Correspondence

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

PCRNet: Point Cloud Registration Network using PointNet Encoding

Point-NAS: A Novel Neural Architecture Search Framework for Point Cloud Analysis.

Deformable Template Network (DTN) for Object Detection

RS-TNet: point cloud transformer with relation-shape awareness for fine-grained 3D visual processing

Auto-Points: Automatic Learning for Point Cloud Analysis with Neural Architecture Search

Structure-aware Shape Correspondence Network for 3D Shape Synthesis.

SCRnet: A Spatial Consistency Guided Network Using Contrastive Learning for Point Cloud Registration

Learning to Train a Point Cloud Reconstruction Network Without Matching

DTF-Net: Category-Level Pose Estimation and Shape Reconstruction via Deformable Template Field

KTNet: Knowledge Transfer for Unpaired 3D Shape Completion

DeTarNet: Decoupling Translation and Rotation by Siamese Network for Point Cloud Registration

HA-TiNet: Learning a Distinctive and General 3D Local Descriptor for Point Cloud Registration

UFO-Net: A Linear Attention-Based Network for Point Cloud Classification