DNN re-init

Re-initialization (re-distancing) of an implicit surface by using a deep neural network. I.e., compute the signed distance function (SDF) to an implicit surface, while preserving the zero level-set.

This code accompanies this JCGT paper.

The code is provided as Python notebooks. They can be run, for example, on Google Colab.

• Figures 2 and 3 correspond to the notebook 'dnn_reinit_1d.ipynb'
• Figure 4 corresponds to the notebook 'dnn_reinit_2d.ipynb'
• Figure 5 corresponds to the notebook 'dnn_reinit_3d.ipynb'
• Figures 1 and 6 correspond to the notebbok 'microstructure_dist.ipynb'

The network architecture, loss, etc are the same (only the dimension and/or model change).

Reference

Link to the JCGT paper where the method is described. The corresponding bibtex entry is

@article{Fayolle2025SDF,
  author =       {Pierre-Alain Fayolle}, 
  title =        {A Zero-Level Set Preserving Technique for Signed Distance Function Computation from an Implicit Surface},
  year =         {2025},
  month =        {May},
  day =          16,
  journal =      {Journal of Computer Graphics Techniques (JCGT)},
  volume =       14,
  number =       1,
  pages =        {185--197},
  url =          {http://jcgt.org/published/0014/01/09/},
  issn =         {2331-7418}
}