Learning Geometry - A Practical Introduction to Neural Distance Fields 

Workshop Description:

Neural implicit representations have recently emerged as a powerful paradigm for modeling complex geometry. Among these, Neural Distance Fields (NDFs) provide a flexible and continuous way to represent shape by learning distance functions with neural networks.
This workshop introduces the fundamentals of neural distance fields, explaining what they are, how they can be obtained through training from point cloud data, and which parameters influence the representation. Participants will explore how architectural and training parameters influence the resulting geometric representation.
The workshop begins with a short introduction to neural distance fields and their role in neural implicit representations. Participants will then train their own neural distance fields using provided software and experiment with different parameters. The session concludes with a group discussion to compare results and reflect on how parameter choices affect the learned geometry.

Organizers: 

• Prof. Dr. Cyrill Stachniss
• Dr.-Ing. Louis Wiesmann

Target Group and Requirements :

The workshop does not require any special knowledge beforehand. A basic understanding of deep learning is helpful but not necessary. Self-developed software will be provided. The neural distance fields will be learned from TLS data. The participants will interact with the software using a graphical user interface (GUI) or command line interface (CLI); no programming skills are needed. The participants can work in groups of 1 to 2 persons.