Approximate 2D-3D Shape Matching for Interactive Applications
Authored by Christoph Petzsch, Paul Roetzer, Zorah Lähner, Florian Bernard
Published in International Conference on 3D Vision (3DV) 2025
Abstract
The problem of matching a 2D contour to deformed version of its 3D counterpart is a challenging setting due to the deformation and dimensionality both heavily impacting point-wise features. In the past, methods were only able to either produce a fast but noisy solution in the product graph or impose higher-order constraints for a smooth but slow solution in the much bigger conjugate product graph. In this work, we propose an approximation of these valuable higher-order terms that are computable in the normal product graph. This leads to an efficient algorithm for high-quality 2D-to-3D matchings and enables novel applications, like an interactive user interface to refine the solution in close-to real-time. We show theoretically that our method is efficient and experimentally that the accuracy gap of our approximation to the optimum is minimal.
Resources
Bibtex
@inproceedings{ petzsch2025approx,
author = { Christoph Petzsch and Paul Roetzer and Zorah Lähner and Florian Bernard },
title = { Approximate 2D-3D Shape Matching for Interactive Applications },
booktitle = { International Conference on 3D Vision (3DV) },
year = { 2025 },
}