A Large-Depth-Range Layer-Based Hologram Dataset for Machine Learning-Based 3D Computer-Generated Holography

TL;DR

The paper addresses the scarcity of large, high-quality hologram datasets for machine-learning-based CGH by introducing KOREATECH-CGH, a public dataset of 6,000 RGB-D and hologram pairs spanning resolutions from $256\times256$ to $2048\times2048$ and a depth range up to 80 mm. It presents AP-LBM, a depth-range-aware hologram-generation method that combines an advanced silhouette-masking layer-based CGH with amplitude projection to improve reconstruction fidelity. A focal-image-projection (FIP) metric is proposed to quantify hologram quality against rendered RGB-D targets, and extensive ML-CGH experiments demonstrate the dataset’s utility for both hologram generation and upscaling. The work enables robust training and benchmarking of next-generation ML-CGH systems for high-depth-range holographic displays and sets the stage for future dynamic-scene and multi-illumination studies.

Abstract

Machine learning-based computer-generated holography (ML-CGH) has advanced rapidly in recent years, yet progress is constrained by the limited availability of high-quality, large-scale hologram datasets. To address this, we present KOREATECH-CGH, a publicly available dataset comprising 6,000 pairs of RGB-D images and complex holograms across resolutions ranging from 256*256 to 2048*2048, with depth ranges extending to the theoretical limits of the angular spectrum method for wide 3D scene coverage. To improve hologram quality at large depth ranges, we introduce amplitude projection, a post-processing technique that replaces amplitude components of hologram wavefields at each depth layer while preserving phase. This approach enhances reconstruction fidelity, achieving 27.01 dB PSNR and 0.87 SSIM, surpassing a recent optimized silhouette-masking layer-based method by 2.03 dB and 0.04 SSIM, respectively. We further validate the utility of KOREATECH-CGH through experiments on hologram generation and super-resolution using state-of-the-art ML models, confirming its applicability for training and evaluating next-generation ML-CGH systems.

Figures (10)

• Figure 1: Overview of our dataset generation pipeline
• Figure 2: The configuration of KOREATECH-CGH consists of (a) RGB and (b) depth map rendered from OptiX, and (c) amplitude and (d) phase of a hologram generated from AP-LBM.
• Figure 3: Object placement in the 3D scene. The figure shows the center positions of the placed objects in the XY-plane and their corresponding depths along the Z-axis.
• Figure 4: Numerical reconstructions of holograms generated using different methods, with focus set to front (left) and back (right) planes.
• Figure 5: Optical setup used for hologram reconstruction. The system includes Lens 1 (focal length: 100 mm) and Lens 2 (focal length: 200 mm), an ND filter (optical density: 3.0), an adjustable slit (OWIS SP60), and a camera lens (Nikon 50 mm), paired with a FLIR Blackfly S camera.