Existing algorithms for computer-generated holography (CGH) rely on iterative approaches that require selecting between computation speed and hologram accuracy. This built-in tradeoff is increasingly incompatible with emerging CGH applications in optogenetics and virtual reality that demand production of high-resolution holograms on short time scales. The present UNC invention is a CGH algorithm based on a non-iterative, unsupervised machine learning model that produces large holograms (up to 11 megavoxels) several orders of magnitude faster than existing techniques and does so with greater accuracy. The unsupervised nature of the model further allows for its tailoring to specific applications through judicious selection of training datasets and loss functions. This DeepCGH approach has been validated in experimental applications.
- DeepCGH: 3D Computer-Generated Holography using Deep Learning Vol. 28, No. 18 / 31 August 2020 / Optics Express 26636