Partial-to-Partial Shape Matching with Geometric Consistency

Viktoria Ehm,Maolin Gao,Paul Roetzer,Marvin Eisenberger,Daniel Cremers,Florian Bernard

2024-05-10

Abstract:Finding correspondences between 3D shapes is an important and long-standing problem in computer vision, graphics and beyond. A prominent challenge are partial-to-partial shape matching settings, which occur when the shapes to match are only observed incompletely (e.g. from 3D scanning). Although partial-to-partial matching is a highly relevant setting in practice, it is rarely explored. Our work bridges the gap between existing (rather artificial) 3D full shape matching and partial-to-partial real-world settings by exploiting geometric consistency as a strong constraint. We demonstrate that it is indeed possible to solve this challenging problem in a variety of settings. For the first time, we achieve geometric consistency for partial-to-partial matching, which is realized by a novel integer non-linear program formalism building on triangle product spaces, along with a new pruning algorithm based on linear integer programming. Further, we generate a new inter-class dataset for partial-to-partial shape-matching. We show that our method outperforms current SOTA methods on both an established intra-class dataset and our novel inter-class dataset.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the partial - to - partial shape matching, which is a very important part in computer vision and graphics. Specifically, when only parts of the shapes to be matched are observed (for example, obtained through 3D scanning), how to find the correspondences between these shapes. This partial - to - partial matching is very common in practical applications, but most of the existing research focuses on the matching of complete shapes or partial - to - complete shape matching, and there is less research on partial - to - partial matching. Therefore, this paper aims to fill this gap and propose a new method to achieve geometrically consistent partial - to - partial shape matching.

The main contributions of the paper include:
1. **Proposing a formal method for geometrically consistent partial - to - partial shape matching**: This method combines the state - of - the - art deep features with the non - linear integer programming method to fuse the advantages of both.
2. **Introducing a pruning search algorithm**: It is used to solve the integer programming problem within a reasonable time.
3. **Providing a new cross - category partial - to - partial shape matching data set**: Based on the SMAL data set, it expands the scope of partial - to - partial shape matching evaluation.
4. **Demonstrating the effectiveness of the method**: On the established within - category data sets and the newly proposed cross - category data sets, the performance of this method is better than the current state - of - the - art methods.

Through these contributions, this paper not only solves the key challenges in partial - to - partial shape matching, but also provides new tools and benchmarks for future related research.