DeepCGH: Computer Generated Holography with Deep Learning
Research at the Pégard lab aims to develop advanced instruments that monitor and manipulate living tissue with light. The team recently developed DeepCGH, a new deep learning algorithm that computes holograms both significantly better and orders of magnitude faster than prior techniques.
Computer Generated Holography (CGH) algorithms calculate how to shape the phase of a laser beam to create custom 3D illumination patterns. This is a difficult, nonlinear, nonconvex, and multidimensional inverse problem for which there is generally no exact solution.
The simplest approach to solve CGH problems relies on iterative algorithms such as Gerchberg-Saxton [1](Check out our introduction to holography lecture), or with direct optimization methods such as NOVO-CGH [2]. However, in all these approaches, the accuracy of holograms improves with the number of iterations and good results requires long computation time.
DeepCGH [3], pushes the performance envelope with both greater speed and accuracy. To accelerate computation, the Pegard Lab relies on convolutional neural network (CNN) that yields holograms without iterations, with an interleaving step that reorganizes the input data to reduce the CNN size. The computation time is fixed and only depends on hologram dimensions. Gains in accuracy are made possible because the CNN is optimized to solve CGH problems for which no ground truth database of exact solutions is available for training, using an unsupervised learning method.
The result is an algorithm that can compute large holograms (~10M Voxels) in milliseconds, and with minimal misplacement of photons, which is critical for applications in biology where suboptimal CGH methods lead to tissue heating.
This work has been made possible by Hossein Eybposh (Ph.D. student), Nicholas Caira (Postdoc), Mathew Atisa (UNC Computer Sciences Major), and Praneeth Chakravarthula (Ph.D. Student) The team received generous support from the Burroughs Wellcome Fund (Career Award at the Scientific Interface, PI: N.Pegard), and from the Nvidia GPU research grant program.
Hossein Eybposh received an OSA Student Paper Award for this work at the 2020 OSA Biophotonics congress).
Mathew Atisa received a first poster prize for his contributions at the 2020 Duke Research Computing Symnposium.
A full summary of this paper, and links to the DeepCGH code can be found here.