“Discontinuity-Aware 2D Neural Fields” by Belhe, Gharbi, Fisher, Georgiev, Ramamoorthi, et al. …
Conference:
Type(s):
Title:
- Discontinuity-Aware 2D Neural Fields
Session/Category Title: Rendering, Neural Fields & Neural Caches
Presenter(s)/Author(s):
Abstract:
Neural image representations offer the possibility of high-fidelity, compact storage, and resolution-independent accuracy, providing an attractive alternative to traditional pixel and grid-based representations. However, coordinate neural networks fail to capture discontinuities present in the image and tend to blur across them; we aim to address this challenge. For many applications, such as representing a resolution-independent rendered image, vector graphics, diffusion curves, or solutions to partial differential equations, we already know the locations of the discontinuities. We take the discontinuity locations as input, represented as linear, quadratic, or cubic Bezier curves, and construct a feature field that is only discontinuous across these locations, and smooth everywhere else. Finally, we use a shallow multi-layer perceptron to decode the features into the signal value. For the feature field construction, we develop a new data structure based on a curved triangular mesh with features stored on the vertices and a subset of the edges of the mesh being marked discontinuous. We show that our method can be used to compress a 100k^2 rendered image into a 25MB file; can be used as a new diffusion curve solver by combining with Monte-Carlo-based methods or directly supervised by the diffusion curve energy; or can be used for compressing 2D physics simulation data.
