Log-GPIS demo

This repository contains a demo code to illustrate the basic idea of our paper Faithful Euclidean Distance Field from Log-Gaussian Process Implicit Surfaces.

Matlab version

R2019b Update 4(9.7.0.1296695)

License

Licensed under GNU General Public License version 3.

2D demo

The observation is a circle to show our Log-GPIS in a 2D case. To show the results, just run LogGPIS_demo_2D.m. Figure 1 to 6 will show the mean distance inference of the whittle kernel and the Matern kernel with lambda varying from 30 to 40. Figure 7 shows the Root Mean Sqrt Error of different kernels and different lambda parameters.

Output using lambda 40 with Whittle and Matern 3/2 kernel respectively looks like (LogGPIS_demo_2D.py):

3D demo

This 3D demo demonstrates that Log-GPIS allows for 3D prediction as a sphere. To see how it goes, just run LogGPIS_demo_3D.m. The result is a black sphere showing the measurements, and the coloured shape or slice is the distance values of the query points.

Citation

If you think Log-GPIS useful in your research, please consider citing our arXiv version, available here:

@article{wu2020faithful,
  title={Faithful Euclidean Distance Field from Log-Gaussian Process Implicit Surfaces},
  author={Wu, Lan and Lee, Ki Myung Brian and Liu, Liyang and Vidal-Calleja, Teresa},
  journal={arXiv preprint arXiv:2010.11487},
  year={2020}
}