A true holographic display doesn’t fake depth by showing each eye a slightly different image. It reproduces the actual wavefront of light an object would give off, so your eyes focus naturally at different depths. It’s the interface that ends the focus problems of flat VR screens. The hard part is computing: to drive the display, software has to calculate a hologram, a pattern that encodes both brightness (amplitude) and timing (phase) of light at every point, and do it fast.
A new SIGGRAPH Asia 2026 Technical Communication, Hologram Representation via Quadratic Phase Gaussian Splatting by Haolong Wang, Yicheng Zhan, Kaan Akşit and Simeng Qiu (submitted Sept 10), builds on one of the more promising recent answers.
From splats to holograms
Gaussian splatting made its name in 3D capture. It builds scenes from millions of soft, fuzzy blobs instead of meshes. Holography researchers noticed that the same idea works for describing a hologram. Last year’s Gaussian Wave Splatting turned splat scenes into holograms, and Zhan and Akşit’s earlier Complex-Valued 2D Gaussian Representation (ECCV 2026) represented the hologram itself as a sum of complex-valued 2D Gaussians. Those Gaussians hold amplitude and phase together and are optimized through a simulation of how light travels. That approach cut the number of parameters to search by up to 10:1 compared with computing every pixel.
What’s new: curved primitives
In that earlier work, each Gaussian’s phase was essentially flat. The new method, CVQPG, gives each primitive a quadratic phase. Think of a small lens rather than a flat tile. Extra learnable parameters control how strongly each one curves.
This matches the physics. A point of light spreading outward produces curved wavefronts, so a primitive that can curve suits the signal it’s trying to describe.
The results, fairly stated
At an equal parameter count, CVQPG beats prior state-of-the-art Gaussian hologram methods by +0.19 dB (RGB) and +0.33 dB (grayscale) on average in reconstructed images. Those are small gains. Nobody would spot 0.19 dB side by side, and this is a short-format Technical Communication, not a full paper.
The more telling result is the frequency analysis. CVQPG preserves the mid-to-high frequency band of natural images, meaning edges, texture and fine detail. That’s where compact hologram representations usually lose quality. The authors’ broader claim is that shaping each primitive’s wavefront is a cheap, lightweight improvement. Getting better quality without adding parameters is exactly what a display needs, since every parameter is computation per frame.
Why it belongs on the interfaces beat
Holographic near-eye displays are one of the few credible paths past today’s headsets, whose lenses fix focus at a single distance. The bottleneck has moved from optics to computation. A compact, differentiable description of a hologram, built from the same primitive the graphics field already uses for capture, is how you get from lab demos toward something wearable. Akşit’s group at UCL has been one of the most consistent contributors on that path. This paper is an incremental step, and a correctly aimed one.
Related Reading
- Hologram Representation via Quadratic Phase Gaussian Splatting — arXiv:2609.11434
- Complex-Valued 2D Gaussian Representation for Computer-Generated Holography — arXiv:2511.15022
- Complex-Valued 2D Gaussian Representation — official code (complight, GitHub)
- Gaussian Wave Splatting for Computer-Generated Holography — project page
- Technical Communications — SIGGRAPH Asia 2026