Real-time 3D Visualization of Radiance Fields on Light Field Displays

This paper tackles a very specific bottleneck: radiance fields look great, but light field displays need many slightly shifted views at once, which normally makes real-time use painfully expensive. The authors propose a shared intermediate representation based on sweeping planes, then composite those planes into the full light-field output in one pass instead of rendering each view separately.

That matters because the method is designed to work across modern scene representations like 3D Gaussian Splatting and Sparse Voxels without retraining. The tradeoff is straightforward: it prioritizes shared, non-directional plane caching for speed, while giving up some fine view-dependent color effects and using a bit more intermediate memory.

The headline numbers are strong for a research prototype: the team reports 200+ FPS at 512p across 45 rendered views on a Looking Glass display, plus up to 22x speedup versus independently rendering each view on standard benchmarks. The paper was submitted on 25 Aug 2025 and revised on 8 Jun 2026.

For game developers, this is less about immediate engine integration and more about where display tech and neural scene rendering are heading. If light field or other multi-view displays become more common in demos, location-based entertainment, or premium hardware, techniques like this could be the difference between a convincing interactive experience and a slideshow.