This code implements the algorithm for the numerical inversion of Square Root Normal Field (SRNF) maps, published in
Hamid Laga, Qian Xie, Ian H Jermyn, and Anuj Srivastava. Numerical inversion of SRNF maps for elastic shape analysis of genus-zero surfaces. IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 39(12), pp.2451-2464 (2017).
The program requires Matlab, and has been extensively tested on Windows. It should also run on Mac. It uses some codes written in C, which need to be compiled with mex. Pre-compiled files for Windows (32 and 64) and Mac (32 and 64) are included.
It is standalone and does not require any setup (apart from having Matlab). It uses some third-party codes, which are included in this distribution with proper license.
This code was developed for research purposes and is not meant for commercial use. We expect that the code may have some bugs and issues. However, while we will be happy to answer some questions, we may not be able to provide support in trouble-shooting the code.
If you use this code, please make sure you cite:
@article{laga2017numerical,
title={Numerical inversion of SRNF maps for elastic shape analysis of genus-zero surfaces},
author={Laga, Hamid and Xie, Qian and Jermyn, Ian H and Srivastava, Anuj},
journal={IEEE transactions on pattern analysis and machine intelligence},
volume={39},
number={12},
pages={2451--2464},
year={2017},
publisher={IEEE}
}and
@article{jermyn2017elastic,
title={Elastic shape analysis of three-dimensional objects},
author={Jermyn, Ian H and Kurtek, Sebastian and Laga, Hamid and Srivastava, Anuj},
journal={Synthesis Lectures on Computer Vision},
volume={12},
number={1},
pages={1--185},
year={2017},
publisher={Morgan \& Claypool Publishers}
}The following scripts are provided to illustrate various usage scenarios. A detailed explanation is provided in the header of each of these scripts.
This code demonstrates how to invert a SRNF map given an initial surface.
This code demonstrates how to invert a SRNF map when no initial surface is available. The optimization startswith a unit sphere. The approach operates on multiscale surfaces, which need to be generated using spherical wavelet transforms (a few example surfaces are included. The code will be made available soon).
The sample code requires harmonic basis of the following resolutions and frequencies (res refers to resolution, LL to frequency):
- res: 8 x 8: LL from 2 to 8
- res: 16 x 16: LL from 8 to 36
- res: 25 x 25: LL of 22, 23 and 36
You may need more if you are dealing with high resolution surfaces.
This script shows how to invert SRNF maps when PCA basis is available for such class of shapes. Note that this one is class-specific.
This script shows how to compute the geodesic between a pair of spatially registered generic surfaces. It uses harmonic basis. The script can also be used for deformation transfer.
This script shows how to compute the geodesic between a pair of spatially registered surfaces when PCA basis are available for that class of shapes. Note that this one is class-specific. In this distribution, we provide PCA basis generated on a subset of the DFAUST dataset.
The script can also be used for deformation transfer.
This script shows how to generate spherical harmonic basis and their derivatives.
This script shows how to generate PCA basis and their derivatives, for a specific class of shapes.
The folder sample_surfaces include exaamples of spherically-parameterized surfaces using this code. However, the original meshes are from TOSCA, SCHREC2007, DFAUST.
Please refer to the file LICENSE. In short, you are free to use the code for research purposes. For any commercial uses, please contact the authors; see below.
Hamid Laga, Sebastian Kurtek, Qian Xie, Ian H. Jermyn, Anuj Srivastava.
For any questions, you can contact [email protected]. Note that the code is provided as it is and we may not be able to provide support on how to use it and fixing potential bugs.