Title: RAVQ-HoloNet: Rate-Adaptive Vector-Quantized Hologram Compression

Abstract

While holography holds immense promise for augmented and virtual reality (AR/VR) technologies, its widespread implementation is hindered by the substantial need for data compression. Current deep learning solutions typically suffer from a lack of rate adaptivity within a single network architecture, often necessitating the deployment of multiple models to accommodate varying bandwidth constraints. To address this, we introduce RAVQ-HoloNet, a framework based on rate-adaptive vector quantization that seamlessly merges rate-adaptive compression with the process of converting image data into phase-only holograms.

RAVQ-HoloNet delivers high-fidelity reconstructions, surpassing existing state-of-the-art techniques through two distinct architectural setups: a standard model designed for low bit rates and a deeper, extended variant optimized for ultra-low bit rate environments. We evaluated these models using the DIV2K dataset as a benchmark for high-fidelity holographic reconstruction. Simulation-based quantitative analysis demonstrates that our approach significantly outperforms current benchmarks. Specifically, in the low bit rate category, our method yields a BD-Rate reduction of -33.91% and a BD-PSNR improvement of 1.02dB compared to the leading method. Furthermore, experiments conducted on an SLM device confirm that our approach enhances image quality and achieves higher contrast.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC