SDFs from Unoriented Point Clouds Using Neural Variational Heat Distances (arxiv.org)

State of the art surface reconstruction - say you took a lidar scan (which gives you a point cloud), to do work with that scan you often want to extract the surface and SDFs are a great way to do represent that

SDF for wave propagation, geometric optical and curl free vector field.The third eikonal assumes that the probabalistic image exists at all points.