SDFConnect: Neural Implicit Surface Reconstruction of a Sparse Point Cloud with Topological Constraints
Abstract
We present a novel approach for neural implicit surface reconstruction from relatively sparse point cloud to ensure the reconstruction of a single connected component. We introduce a topological loss term based on persistent homology to reconstruct a manifold object of genus 1. Building on the Neural Pull [25] framework, our method demonstrates superior performance in preserving the integrity of complex 3D geometries, evident through both visual and empirical comparisons. Our contributions include the integration of persistent diagrams to refine shape topology and a topological loss term to constrain existing reconstruction pipelines to a single connected component. This advancement allows for the seamless integration of topological data analysis with implicit surface reconstruction.
Cite
Text
Jignasu et al. "SDFConnect: Neural Implicit Surface Reconstruction of a Sparse Point Cloud with Topological Constraints." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2024. doi:10.1109/CVPRW63382.2024.00536Markdown
[Jignasu et al. "SDFConnect: Neural Implicit Surface Reconstruction of a Sparse Point Cloud with Topological Constraints." IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, 2024.](https://mlanthology.org/cvprw/2024/jignasu2024cvprw-sdfconnect/) doi:10.1109/CVPRW63382.2024.00536BibTeX
@inproceedings{jignasu2024cvprw-sdfconnect,
title = {{SDFConnect: Neural Implicit Surface Reconstruction of a Sparse Point Cloud with Topological Constraints}},
author = {Jignasu, Anushrut and Balu, Aditya and Sarkar, Soumik and Hegde, Chinmay and Ganapathysubramanian, Baskar and Krishnamurthy, Adarsh},
booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops},
year = {2024},
pages = {5271-5279},
doi = {10.1109/CVPRW63382.2024.00536},
url = {https://mlanthology.org/cvprw/2024/jignasu2024cvprw-sdfconnect/}
}