Complex-Valued 2D Gaussian Representation for Computer-Generated Holography

2025-11-19

AI Summary

To address the challenges of large parameter search spaces, high memory overhead, low reconstruction fidelity, and poor scalability in computer-generated holography (CGH), this paper proposes a parametric hologram representation based on structured complex-valued 2D Gaussian primitives—replacing conventional pixel-wise storage. We integrate differentiable rasterization with a GPU-accelerated free-space light propagation kernel to construct an end-to-end trainable framework, augmented by phase-constrained conversion and gradient-based optimization strategies. Our approach drastically reduces parameter dimensionality, suppresses reconstruction noise, and enhances both optimization efficiency and image quality. Experiments demonstrate a 70% reduction in GPU memory consumption, a 50% acceleration in optimization speed, and superior reconstruction fidelity compared to state-of-the-art methods. Moreover, the framework is compatible with multiple practical holographic formats, ensuring broad applicability.

Technology Category

• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Non-convex Optimization
• Machine Learning: Optimization

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications
• Security and Privacy: Applications of cryptography

Abstract

We propose a new hologram representation based on structured complex-valued 2D Gaussian primitives, which replaces per-pixel information storage and reduces the parameter search space by up to 10:1. To enable end-to-end training, we develop a differentiable rasterizer for our representation, integrated with a GPU-optimized light propagation kernel in free space. Our extensive experiments show that our method achieves up to 2.5x lower VRAM usage and 50% faster optimization while producing higher-fidelity reconstructions than existing methods. We further introduce a conversion procedure that adapts our representation to practical hologram formats, including smooth and random phase-only holograms. Our experiments show that this procedure can effectively suppress noise artifacts observed in previous methods. By reducing the hologram parameter search space, our representation enables a more scalable hologram estimation in the next-generation computer-generated holography systems.

Problem

• Proposes complex-valued 2D Gaussian primitives to replace pixel-based hologram storage
• Reduces hologram parameter search space by 10:1 for scalable hologram estimation
• Enables higher-fidelity reconstructions with lower VRAM usage and faster optimization

Innovation

• Complex-valued 2D Gaussian primitives replace pixel storage
• Differentiable rasterizer enables end-to-end GPU training
• Conversion procedure adapts representation to phase-only holograms