Generative models based on eigendecomposition for dense ray tracing

Abstract

In this work, we present an algorithm capable of emulating ray trajectories that obey the least action principle. The method is based on spectral decomposition of geometric shapes taken from a set of raypaths. As the proposed work relies on shape analysis, it is agnostic on the underlying physics of raypath generation. As such, it is independent of the ray tracing method used to generate the training paths. In cases of mildly heterogeneous media or scenarios with a limited number of geometrical scatters, we show that the algorithm is capable of efficiently populating a given scenario with a dense array of emulated rays whose trajectories are in close agreement with actual rays. We argue that the algorithm also serves as an effective method capable of detecting regions where ray variation is high, such as when possible shadow zones are present.