Geometry Processing with Neural Fields | Guandao Yang

Описание: Join the Learning on Graphs and Geometry Reading Group: https://hannes-stark.com/logag-readin...

Paper “Geometry Processing with Neural Fields”: http://vladlen.info/papers/neural-fie...

Abstract: Most existing geometry processing algorithms use meshes as the default shape representation. Manipulating meshes, however, requires one to maintain high quality in the surface discretization. For example, changing the topology of a mesh usually requires additional procedures such as remeshing. This paper instead proposes the use of neural fields for geometry processing. Neural fields can compactly store complicated shapes without spatial discretization. Moreover, neural fields are infinitely differentiable, which allows them to be optimized for objectives that involve higher-order derivatives. This raises the question: can geometry processing be done entirely using neural fields? We introduce loss functions and architectures to show that some of the most challenging geometry processing tasks, such as deformation and filtering, can be done with neural fields. Experimental results show that our methods are on par with the well-established mesh-based methods without committing to a particular surface discretization. Code is available at https://github.com/stevenygd/NFGP.

Authors: Guandao Yang, Serge Belongie, Bharath Hariharan, Vladlen Koltun

Twitter Hannes:   / hannesstaerk  
Twitter Dominique:   / dom_beaini  
Twitter Valence Discovery:   / valence_ai  

Reading Group Slack: https://join.slack.com/t/logag/shared...

~

00:00 Begin
00:14 Intro & Presentation Overview
02:45 What is Geometric Processing?
06:10 Implicit Fields
09:36 Training Neural Fields
13:56 Geometry Processing Using Neural Fields
22:26 Deformation with Neural Fields
54:14 Deformation Results
59:58 Limitations and Future Works
1:07:24 Preservation of Field Properties
1:10:04 Q&A
1:24:12 Conclusion