Friday, May 2025

11:40 AM - 12:00 PM

Room: 220C

Session: Artificial Intelligence for AR/VR/MR

Filter-Free 3D HoloNet with Hardware-Aware Calibration

Invited

• Yifan Peng

  University of Hong Kong

Description:

Computational holographic displays often depend on iterative CGH algorithms and bulky physical filters to achieve high-quality image reconstruction, making them both time-consuming and impractical. This trade-off between speed and image quality becomes even more challenging for 3D holographic imagery. This work introduces 3D-HoloNet, a deep learning-based CGH algorithm that generates phase-only holograms for 3D scenes (represented as RGB-D images) in real time. By incorporating a learned, camera-calibrated wave propagation model and a phase regularization prior, 3D-HoloNet effectively handles unfiltered holographic display setups, even when impacted by hardware imperfections. Tests on such setups demonstrate that 3D-HoloNet achieves 30 fps at full HD resolution for a single-color channel using a consumer GPU, while maintaining image quality comparable to iterative methods across multiple focal distances.