Focal Surface Holographic Light Transport using Learned Spatially Adaptive Convolutions

TL;DR

This work replaces multiple planes with a focal surface and introduces a learned light transport model that could propagate a light field from a source plane to the focal surface in a single inference.

Abstract

Computer-Generated Holography (CGH) is a set of algorithmic methods for identifying holograms that reconstruct Three-Dimensional (3D) scenes in holographic displays. CGH algorithms decompose 3D scenes into multiplanes at different depth levels and rely on simulations of light that propagated from a source plane to a targeted plane. Thus, for n planes, CGH typically optimizes holograms using n plane-to-plane light transport simulations, leading to major time and computational demands. Our work replaces multiple planes with a focal surface and introduces a learned light transport model that could propagate a light field from a source plane to the focal surface in a single inference. Our learned light transport model leverages spatially adaptive convolution to achieve depth-varying propagation demanded by targeted focal surfaces. The proposed model reduces the hologram optimization process up to 1.5x, which contributes to hologram dataset generation and the training of future learned CGH models.

Figures (5)

• Figure 1: Conventional Light Transport VS. Proposed Focal Surface Light Transport.(Source image: Tobi 87, Link: https://commons.wikimedia.org/wiki/File:Calanque_d'En_Vau-Cassis.jpg)
• Figure 2: Our proposed learned focal surface light transport model. The process starts with an input hologram $\mathbf{H}$ and a focal surface $\mathbf{D}$ to generate spatially varying kernels $[\mathbf{V_{i}}]$, where $i = 0, 1, 2, 3$ indicates the index of scales. Those kernels are utilized in the Spatially Adaptive Module (SAM) to achieve focal surface light transport. In the SAM, $\mathbf{V}_{3}^{0}, \mathbf{V}_{3}^{j}, \mathbf{V}_{3}^{l}, \mathbf{V}_{3}^{z}$ represent kernels used at different spatial locations, where $0$, $j$, $l$, and $z$ indicate specific positions. (Source image: Tobi 87, Link: https://commons.wikimedia.org/wiki/File:Calanque_d'En_Vau-Cassis.jpg)
• Figure 3: Visual comparison of simulating light transported onto a focal surface (specified in the first row of each case) at 0 mm and 10 mm propagation distances. The ground truth is obtained via ASM matsushima2009band. Both focused and defocused regions indicate poor performance of the U-Net model. (Source image: Matt H. Wade, Link: https://commons.wikimedia.org/wiki/File:Cinderella_Castle_2013_Wade.jpg)
• Figure 4: Visual comparison on simulated holograms optimized using ASM 6 and Ours 6 under 0 mm propagation distance. All holograms are reconstructed using ASM for evaluation. (Source image : Jaimie Phillips, Link: https://commons.wikimedia.org/wiki/File:Dewdrops_on_leaves_(Unsplash).jpg)
• Figure 5: Comparing experimental captures of ASM 6 and Ours 6 under 0 mm propagation distances. (Source image : Jaimie Phillips, Link: https://commons.wikimedia.org/wiki/File:Dewdrops_on_leaves_(Unsplash).jpg)