Deep Manifold Prior
Authors:
- 1 October 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1107-1116
Abstract
We present a prior for manifold structured data, such as surfaces of 3D shapes, where deep neural networks are adopted to reconstruct a target shape using gradient descent starting from a random initialization. We show that surfaces generated this way are smooth, with limiting behavior characterized by Gaussian processes, and we mathematically derive such properties for fully-connected as well as convolutional networks. We demonstrate our method in a variety of manifold reconstruction applications, such as point cloud denoising and interpolation, achieving considerably better results against competitive baselines while requiring no training data. We also show that when training data is available, our method allows developing alternate parametrizations of surfaces under the framework of Atlas-Net [14], leading to a compact network architecture and better reconstruction results on standard image to shape reconstruction benchmarks.Keywords
This publication has 26 references indexed in Scilit:
- A Point Set Generation Network for 3D Object Reconstruction from a Single Image2017
- 3D-R2N2: A Unified Approach for Single and Multi-view 3D Object Reconstruction2016
- Multi-view 3D Models from Single Images with a Convolutional Network2016
- Screened poisson surface reconstructionACM Transactions on Graphics, 2013
- Feature Preserving Point Set Surfaces based on Non‐Linear Kernel RegressionComputer Graphics Forum, 2009
- Computing and rendering point set surfacesIEEE Transactions on Visualization and Computer Graphics, 2003
- Nonlinear Dimensionality Reduction by Locally Linear EmbeddingScience, 2000
- A Global Geometric Framework for Nonlinear Dimensionality ReductionScience, 2000
- Orbits of Families of Vector Fields and Integrability of DistributionsTransactions of the American Mathematical Society, 1973
- A two-dimensional interpolation function for irregularly-spaced data1968