This is the authors' abstract. We don't add key points for this paper.
Deep learning has shown significant promise for computer-generated holography (CGH), particularly in enabling real-time rendering. However, conventional U-Net-based methods exhibit critical limitations in computational resource efficiency and feature extraction capability, compromising both reconstruction quality and processing efficiency. To overcome this issue, this paper proposes a complex-valued efficient hybrid attention network (CEHAN) for high-quality hologram generation. The architecture comprises two specialized sub-networks: a complex amplitude inference network (CAIN) and a hologram encoding network (HEN). To enhance both reconstruction accuracy and computational efficiency, a complex efficient attention (CEA) mechanism is incorporated into the downsampling module. Furthermore, a hybrid attention block (HAB) integrates both channel and spatial attention mechanisms to optimize feature extraction. The proposed approach achieves a computational time of 16 ms per frame, while attaining an average PSNR of 35.71 dB and an SSIM of 0.944 on the DIV2K dataset, surpassing conventional methods. Numerical simulations and optical experiments demonstrate that the proposed method achieves superior detail reproduction and enhanced image quality while maintaining reduced computational demands. These results underscore the framework's strong potential for practical applications in holographic displays.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: