ImagingOpt.jl

Author gaurav-arya
Github
Popularity
6 Stars
Updated Last
5 Months Ago
Started In
April 2022

ImagingOpt

Prerequisites

Clone WavePropagation.jl, ImplicitAdjoints.jl and this repo and run

dev {path to WavePropagation.jl}
dev {path to ImplicitAdjoints.jl}
dev {path to this repo}

in Julia's package manager.

Directory structure

  • Code is in src
    • src/forward.jl defines the linear operator G which convolves with all the PSFs.
    • src/pipeline.jl makes a differentiable pipeline using the physics functions from WavePropagation.jl, the Lasso reconstruction function from ImplicitAdjoints.jl, and the linear operator from src/forward.jl. Everything is composed together in the function loss.
    • src/prepare.jl defines the format of all the input parameters and has utility functions for doing all the precomputation required before optimization.
    • src/optimize.jl provides some functions for running an optimization given the name of a JSON file with parameters.
  • Parameter files describing an imaging optimization problem are expected in params.
  • Surrogate data (generated from Python) is expected in surdata.
  • Optimization data is saved in optdata.

Example

A toy example small enough to easily run on a laptop:

using ImagingOpt
using Plots

optname = run_opt("2dsmall")

smalls = get_smalls(optname) # concise info about each iteration
MSEs = [small["res"].MSE for small in smalls]
plot(MSEs, yaxis=:log, dpi=300)

raw = get_raw(optname) # all the raw data (psfs, etc.) from a single iteration (default is last iteration).
PSF = raw["res"].PSFs[:,:,1,1,1] # only 1 channel so only 1 PSF (indices are x,y,depth,frequency,configuration)
heatmap(PSF, aspect_ratio=:equal, size=(400,400), dpi=300)

MSEs PSF

Citation

@misc{https://doi.org/10.48550/arxiv.2201.12348,
  doi = {10.48550/ARXIV.2201.12348},
  url = {https://arxiv.org/abs/2201.12348},
  author = {Arya, Gaurav and Li, William F. and Roques-Carmes, Charles and Solja\v{c}i\'{c}, Marin and Johnson, Steven G. and Lin, Zin},
  keywords = {Image and Video Processing (eess.IV), Optimization and Control (math.OC), Optics (physics.optics), FOS: Electrical engineering, electronic engineering, information engineering, FOS: Electrical engineering, electronic engineering, information engineering, FOS: Mathematics, FOS: Mathematics, FOS: Physical sciences, FOS: Physical sciences},
  title = {End-to-End Optimization of Metasurfaces for Imaging with Compressed Sensing},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}