Abstract:Finding correspondences between 3D shapes is an important and long-standing problem in computer vision, graphics and beyond. While approaches based on machine learning dominate modern 3D shape matching, almost all existing (learning-based) methods require that at least one of the involved shapes is complete. In contrast, the most challenging and arguably most practically relevant setting of matching partially observed shapes, is currently underexplored. One important factor is that existing datasets contain only a small number of shapes (typically below 100), which are unable to serve data-hungry machine learning approaches, particularly in the unsupervised regime. In addition, the type of partiality present in existing datasets is often artificial and far from realistic. To address these limitations and to encourage research on these relevant settings, we provide a generic and flexible framework for the procedural generation of challenging partial shape matching scenarios. Our framework allows for a virtually infinite generation of partial shape matching instances from a finite set of shapes with complete geometry. Further, we manually create cross-dataset correspondences between seven existing (complete geometry) shape matching datasets, leading to a total of 2543 shapes. Based on this, we propose several challenging partial benchmark settings, for which we evaluate respective state-of-the-art methods as baselines.

What problem does this paper attempt to address?

### Problems the paper attempts to solve The paper attempts to address the challenges in **partial shape matching**. Specifically, most of the existing machine - learning - based methods for partial shape matching require at least one complete shape. However, in practical applications, shapes with missing parts are often encountered due to occlusion, limited sensor coverage, or errors during 3D scanning. In such cases, partial shape matching becomes very difficult and is of great practical significance. The main contributions of the paper are as follows: 1. **Generation framework**: A flexible and extensible framework is proposed, which can generate almost unlimited training examples based on existing complete or partial geometric data for partial shape - matching tasks. This framework supports the selection and propagation of cross - category matching and provides a versatile solution for creating customized training data to meet specific application requirements. 2. **Benchmark dataset**: A challenging partial non - rigid shape dataset named BeCoS is constructed, which is a carefully curated instance of the existing framework. This dataset serves as a standard benchmark for shape - matching methods, promoting the development of more robust and accurate algorithms. 3. **Performance analysis**: Through extensive experiments, the performance of the current state - of - the - art methods on the new benchmark is analyzed, revealing the limitations and key challenges of these methods and emphasizing the need for further research in the field of partial shape matching. ### Background and motivation Partial shape matching is a long - standing problem in computer vision and graphics, with a wide range of applications, such as texture and deformation transfer, statistical shape analysis, etc. Although machine - learning - based methods dominate modern 3D shape matching, they usually require at least one complete shape. This limits the application of these methods in real - world scenarios because shapes in the real world often have missing parts. The existing partial shape - matching datasets have the following problems: - **Limited quantity**: Most existing datasets contain a small number of shapes (usually less than 100), which cannot meet the needs of data - driven methods, especially in an unsupervised setting. - **Unrealistic partiality**: The way partial shapes are generated in existing datasets is often artificial and quite different from the actual situation. - **Lack of diversity**: Existing datasets lack diversity in shape categories, affecting the generalization ability of methods. To solve these problems, the authors propose a generation framework that can generate a large number of real and diverse partial shape - matching instances and construct a new benchmark dataset BeCoS to promote the research progress in the field of partial shape matching.

Beyond Complete Shapes: A quantitative Evaluation of 3D Shape Matching Algorithms

3D Shape Matching Using Collinearity Constraint

Partial-to-Partial Shape Matching with Geometric Consistency

A Comparison of 3D Shape Retrieval Methods Based on a Large-Scale Benchmark Supporting Multimodal Queries

Full and Partial Shape Similarity Through Sparse Descriptor Reconstruction

Isometric Multi-Shape Matching

On Unsupervised Partial Shape Correspondence

Shape Topics: A Compact Representation and New Algorithms for 3D Partial Shape Retrieval

DiscoMatch: Fast Discrete Optimisation for Geometrically Consistent 3D Shape Matching

3D shape representation, matching and recognition

Zero-Shot 3D Shape Correspondence

Automatic 3-D grayscale volume matching and shape analysis

Distances and Algorithms to Compare Sets of Shapes for Automated Biological Morphometrics

Mismatched: Evaluating the Limits of Image Matching Approaches and Benchmarks

Wormhole Loss for Partial Shape Matching

A Point Pairing Method for 3D Partial Shape Matching

A Survey on Partial Retrieval of 3D Shapes

Shape Similarity Assessment Based On Partial Feature Aggregation And Ranking Lists

The Generalized Shape Distributions for Shape Matching and Analysis

3D Shape Completion on Unseen Categories:A Weakly-supervised Approach

Shape Complementarity Analysis for Objects of Arbitrary Shape