Structurally guided wavefront neural encoder for broad-range generalizable 3D holography
In the authors' words
Abstract Deep learning has advanced computer-generated holography, but learned models remain coupled to fixed wavelengths and distances, limiting generalization. Here we present the structurally guided wavefront neural encoder (SGWN-Encoder), which integrates diffraction physics with neural networks and employs a structural propagation strategy that analytically strips the parameter-sensitive global phase. This reveals a scale-normalized description of diffraction, letting the network learn transferable light-field evolution rules and generalize across propagation distances and visible wavelengths by direct inference. A dataset-free paradigm built on statistically diverse random fields teaches the model the structural mapping from complex amplitude to phase-only holograms instead of external image distributions. Without fine-tuning, a single SGWN-Encoder maintains a distance-averaged RGB PSNR of 31.9 dB and SSIM of 0.91 over the primary 2–200 mm range, retaining reconstruction up to 500 mm. Numerical and optical experiments on full-color 3D scenes establish a physically grounded route toward adaptive holographic displays.
In plain words AI summary
Current deep learning models for holography are tied to specific wavelengths and distances, which restricts their use. The authors created the SGWN-Encoder, which combines physics of light diffraction with neural networks. This approach removes phase parameters that depend on distance and wavelength, allowing the model to work across different conditions. The model was trained using random fields rather than real images, learning to map light amplitude to hologram phase patterns. Without adjustment, one model achieved 31.9 dB PSNR and 0.91 SSIM across 2–200 mm distances, working up to 500 mm.
Written by Kimi K2.5 using only the abstract above. Jev checked that every sentence is supported by the abstract sentences it cites, and removed any that were not. Point at a sentence to highlight its source.
Appeared: Wednesday, November 18. Nature Communications. Peer-reviewed journal.