Abstract:In this paper, we propose a learning-based framework for non-rigid shape registration without correspondence supervision. Traditional shape registration techniques typically rely on correspondences induced by extrinsic proximity, therefore can fail in the presence of large intrinsic deformations. Spectral mapping methods overcome this challenge by embedding shapes into, geometric or learned, high-dimensional spaces, where shapes are easier to align. However, due to the dependency on abstract, non-linear embedding schemes, the latter can be vulnerable with respect to perturbed or alien input. In light of this, our framework takes the best of both worlds. Namely, we deform source mesh towards the target point cloud, guided by correspondences induced by high-dimensional embeddings learned from deep functional maps (DFM). In particular, the correspondences are dynamically updated according to the intermediate registrations and filtered by consistency prior, which prominently robustify the overall pipeline. Moreover, in order to alleviate the requirement of extrinsically aligned input, we train an orientation regressor on a set of aligned synthetic shapes independent of the training shapes for DFM. Empirical results show that, with as few as dozens of training shapes of limited variability, our pipeline achieves state-of-the-art results on several benchmarks of non-rigid point cloud matching, but also delivers high-quality correspondences between unseen challenging shape pairs that undergo both significant extrinsic and intrinsic deformations, in which case neither traditional registration methods nor intrinsic methods work. The code is available at <a class="link-external link-https" href="https://github.com/rqhuang88/DFR" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve This paper aims to address the challenges in non-rigid shape registration, particularly in registering point clouds without correspondence supervision. Traditional methods often rely on external proximity-induced correspondences, which tend to fail in the presence of significant intrinsic deformations. Although spectral mapping methods overcome this challenge by embedding shapes into high-dimensional space, they are highly sensitive to perturbations or unfamiliar inputs. To tackle these issues, the authors propose a framework that combines classical shape registration with deep learning-based embedding techniques. ### Specific Problem Description 1. **Limitations of Traditional Methods**: - **Reliance on External Proximity**: Traditional methods typically rely on external proximity between point clouds to establish correspondences, which can easily fail in the presence of significant intrinsic deformations. - **Fragility of High-Dimensional Embedding**: While spectral mapping methods can better handle intrinsic deformations, they are very sensitive to perturbations or unfamiliar inputs. 2. **Objectives**: - **Non-Rigid Shape Registration without Correspondence Supervision**: Achieve non-rigid shape registration between point clouds without correspondence supervision. - **Robustness and Generalization**: The proposed method needs to be robust and generalize well when dealing with significant deformations and heterogeneous shapes. ### Solution 1. **Hybrid Framework**: - **Deep Functional Maps (DFM)**: Utilize DFM to learn high-dimensional embeddings of point clouds, dynamically update correspondences, and filter them through consistency priors. - **Shape Deformation**: Use the correspondences learned by DFM to guide the deformation of the source mesh towards the target point cloud, maintaining similarity in both ambient space and high-dimensional embedding space. 2. **Key Design**: - **Feature Extractor**: Train a point feature extractor, using DFM as a teacher network, to generate features through self-supervised learning. - **Orientation Regressor**: Train an orientation regressor to automatically preprocess input point clouds from arbitrary orientations. - **Dynamic Correspondence Update**: Dynamically update correspondences during optimization to ensure their accuracy. - **Two-Stage Registration**: The first stage performs registration based on learned features, and the second stage fine-tunes based on coordinates to improve registration quality. ### Experimental Results 1. **Near-Isometric Benchmark**: - In the presence of significant deformations, this method provides high-quality deformation results compared to traditional methods like NDP and AMM. - In multiple test cases, this method performs best in 5 cases, especially in the most challenging tasks such as FAUST_r/SHREC19_r and FAUST_r/SHREC19_r. 2. **Non-Isometric Benchmark**: - Tested on challenging non-isometric datasets like SHREC07-H and DT4D-H, which contain shapes not seen during training. - The method excels in handling heterogeneous shapes and significant deformations, particularly in the SHREC07-H dataset, capable of dealing with heterogeneous human shapes with vertex counts ranging from 3000 to 15000, and is robust to the choice of template shapes. ### Summary This paper proposes a hybrid framework that combines classical shape registration with deep learning-based embedding techniques, achieving non-rigid shape registration between point clouds without correspondence supervision. Experimental results demonstrate that the method is robust and generalizes well when handling significant deformations and heterogeneous shapes, outperforming several existing methods.

Non-Rigid Shape Registration via Deep Functional Maps Prior

Multiway Non-Rigid Point Cloud Registration Via Learned Functional Map Synchronization

Non-rigid 3D Shape Registration using an Adaptive Template

Coherent Point Drift Revisited for Non-rigid Shape Matching and Registration

Bidirectional Patch-based Correlations with Local Rigidity for Global Nonrigid Registration

Deep Correspondence Matching-Based Robust Point Cloud Registration of Profiled Parts

DeepASDM: a Deep Learning Framework for Affine and Deformable Image Registration Incorporating a Statistical Deformation Model

Non-rigid Registration in 3D Implicit Vector Space

Sparse Non-rigid Registration of 3D Shapes.

Spatially and Spectrally Consistent Deep Functional Maps

Deep Weighted Consensus: Dense correspondence confidence maps for 3D shape registration

A Robust Nonrigid Point Set Registration Method Based on Collaborative Correspondences

Deep Graph Matching Based Dense Correspondence Learning Between Non-Rigid Point Clouds.

An Efficient Volumetric Method for Non-Rigid Registration

Fast and Robust Non-Rigid Registration Using Accelerated Majorization-Minimization

Deep Frequency-Aware Functional Maps for Robust Shape Matching

Equivariant Local Reference Frames for Unsupervised Non-rigid Point Cloud Shape Correspondence

Self-supervised Learning of Implicit Shape Representation with Dense Correspondence for Deformable Objects

A Unified Framework For Nonrigid Point Set Registration Via Coregularized Least Squares

Neural Intrinsic Embedding for Non-rigid Point Cloud Matching

Deformation-Guided Unsupervised Non-Rigid Shape Matching