Please note: This PhD seminar will take place online.
Wenyou Wang, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Toshiya Hachisuka
This work introduces a new spatio-directional neural encoding that is compact and efficient, and supports all-frequency signals in both space and direction. Current learnable encodings focus on Cartesian orthonormal spaces, which have been shown to be useful for representing high-frequency signals in the spatial domain. However, directly applying these encodings in the directional domain results in distortions, singularities, and discontinuities. As a result, most related works have used more traditional encodings for the directional domain, which lack the expressivity of learnable neural encodings.
To address this limitation, the authors propose a new angular encoding that generalizes the hash-grid approach of Müller et al. [2022] to the directional domain by encoding directions using a hierarchical geodesic grid. Each vertex in the geodesic grid stores a learnable latent parameter, which is used as input to a neural network. Building upon this directional encoding, the authors further introduce a five-dimensional encoding for spatio-directional signals. Their results demonstrate that both encodings significantly outperform other hash-based alternatives. The authors also apply their five-dimensional encoding to neural path guiding, achieving up to a 2× reduction in variance compared to the state of the art for the same number of samples.