Variable Bitrate Neural Fields
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
DescriptionNeural fields have emerged as a new representation for graphics. State-of-the-art architectures use trainable feature grids that allow the use of efficient neural networks at the cost of memory. Our dictionary method allows us to learn compressed discrete feature grids which reduces memory by up to 100x and enables streaming.