Physics Informed Neural Fields for Smoke Reconstruction With Sparse Data
Event Type
Technical Paper
Interest Areas
Research & Education
Presentation Types
Registration Categories
Full Conference Supporter
Full Conference
Virtual Conference Supporter
Virtual Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
DescriptionWe present a novel method to reconstruct continuous fluid fields from sparse video frames with unknown lighting conditions by leveraging the governing physics (i.e., Navier-Stokes equations) in an end-to-end optimization. Throughout our hybrid architecture, fluid interactions encompassing static obstacles are reconstructed for the first time without requiring geometry information.