Assessing Learned Models for Phase-only Hologram Compression

2025-07-09

AI Summary

This study addresses the challenge of efficient compression for pure-phase holograms. We systematically evaluate four implicit neural representation (INR)–based variational autoencoders—MLP, SIREN, FilmSIREN, and TAESD—on phase-only holographic data. Experimental results show that pre-trained image VAEs (e.g., TAESD) exhibit poor generalization to holographic phase distributions, failing to simultaneously achieve low bitrates and high-fidelity reconstruction. In contrast, SIREN—a sinusoidal activation-based INR specifically designed for oscillatory signals—achieves a 40% compression ratio and 34.54 dB PSNR using only 4.9k parameters, outperforming all alternatives. Our analysis reveals the strong dependence of hologram compression on model-inductive bias: lightweight, phase-domain–tailored INRs significantly surpass transferred image VAEs in both efficiency and fidelity. This work establishes a novel end-to-end compression paradigm for holographic display, emphasizing task-specific architectural design over generic deep generative models.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Representation Learning for Vision
• Data Mining & Knowledge Management: Data Compression

Application Category

• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

We evaluate the performance of four common learned models utilizing INR and VAE structures for compressing phase-only holograms in holographic displays. The evaluated models include a vanilla MLP, SIREN, and FilmSIREN, with TAESD as the representative VAE model. Our experiments reveal that a pretrained image VAE, TAESD, with 2.2M parameters struggles with phase-only hologram compression, revealing the need for task-specific adaptations. Among the INRs, SIREN with 4.9k parameters achieves %40 compression with high quality in the reconstructed 3D images (PSNR = 34.54 dB). These results emphasize the effectiveness of INRs and identify the limitations of pretrained image compression VAEs for hologram compression task.

Problem

• Evaluate learned models for phase-only hologram compression
• Compare INR and VAE structures in holographic displays
• Identify limitations of pretrained VAEs for hologram compression

Innovation

• Evaluates INR and VAE models for hologram compression
• SIREN achieves 40% compression with high quality
• Pretrained VAEs need task-specific adaptations for holograms