Complex-Valued Holographic Radiance Fields

Abstract

Modeling wave properties of light is an important milestone for advancing physically-based rendering. In this paper, we propose complex-valued holographic radiance fields, a method that optimizes scenes without relying on intensity-based intermediaries. By leveraging multi-view images, our method directly optimizes a scene representation using complex-valued Gaussian primitives representing amplitude and phase values aligned with the scene geometry. Our approach eliminates the need for computationally expensive holographic rendering that typically utilizes a single view of a given scene. This accelerates holographic rendering speed by 30x-10,000x while achieving on-par image quality with state-of-the-art holography methods, representing a promising step towards bridging the representation gap between modeling wave properties of light and 3D geometry of scenes.

Figures (24)

• Figure 1: Left: Existing methods do not preserve geometry of scenes under viewpoint changes, necessitating transformation or recalculation per-view. Right: Our complex-valued holographic radiance field offers a 3D consistent representation in contrast.
• Figure 2: Without the need to query an intensity-based radiance field, our approach models a complex-valued holographic radiance field using Gaussian primitives with intrinsic amplitude and phase properties, enabling scene geometry-aware amplitude and phase modeling across viewpoints and efficient rendering through a differentiable multi-layer propagation pipeline.
• Figure 3: Comparison of different hologram synthesis methods across different viewpoints in simulation. The top row shows our method with scene geometry-aware representations from large left to right rotations for novel views. The middle rows show the existing methods that rely on intensity-based intermediaries, which fail to maintain consistency across novel views. The bottom row shows the intensity ground truth. * We reimplement choi2025gaussian to demonstrate the results of ; The original non-rotational result shows higher image quality and different defocus blur than our reimplementation.
• Figure 4: Performance and memory usage analysis. (a) Render time vs resolution for different Gaussian counts under 1 plane. (b) Memory usage & render time vs number of depth planes. (c) Memory usage & render time vs number of Gaussians. (d) Memory usage vs resolution for different numbers of Gaussians.
• Figure 5: Motion parallax across datasets: NeRF Synthetic (left) shows a controlled scene with minimal parallax; LLFF (center) shows an indoor scene with vast-moving foreground motion (red region); Mip-NeRF 360 (right) shows an in-the-wild scene with wide background shifts (red region). Regions exhibiting moderate motion parallax that does not cause phase discontinuities in our experiments are marked in green.