1.5. Configuration file

Configuration files are YAML human-readable files. Note that some of them are optional and can be absent from the configuration file. If this is the case, a default value will be used.

1.5.1. Spatially invariant

telescope:
    diameter: 100.0
    central_obscuration : 0.0
    spider : 0

images:
    n_pixel : 64
    pix_size : 0.059
    apodization_border : 6

object1:
    wavelength : 8542.0
    image_filter: tophat
    cutoff : [0.75, 0.80]

object2:
    wavelength : 8542.0
    image_filter: tophat
    cutoff : [0.75, 0.80]

optimization:
    gpu : 0
    transform : softplus
    softplus_scale : 1.0
    lr_obj : 0.02
    lr_modes : 0.08
    show_object_info: False

regularization:
    iuwt1:
        variable : object
        lambda : 0.0
        nbands : 5

psf:
    model : kl
    nmax_modes : 44

initialization:
    object : contrast
    modes_std : 0.0

annealing:
    type: sigmoid
    start_pct : 0.0
    end_pct : 0.85

1.5.1.1. telescope

  • diameter: diameter of the telescope in cm

  • central_obscuration : central obscuration of the telescope in cm

  • spider : width of the spider in cm

1.5.1.2. images

  • n_pixel : size in pixels of the images to be deconvolved. In case the images are mosaicked to take into account the size of the anisoplanatic patch, this keyword gives the size of the patches

  • pix_size : pixel size in arcsec

  • apodization_border : border of the apodization in pixels to avoid edge effects in the Fourier transform used for the convolution

1.5.1.3. objects

Since several objects can be deconvolved simultaneously, one needs to define the parameters for each object. Objects are defined as objectx, where x is an index starting from 1.

  • wavelength : wavelength of the object in angstroms

  • image_filter : Fourier filter to be used for the image. Options are tophat (a simple tophat filter to avoid frequencies above the cutoff) or scharmer (the filter from Lofdahl & Scharmer 1994).

  • cutoff : cutoff frequency. The first element indicates the cutoff frequency below which the filter is 1. The second element marks the frequency above which the filter is 0. In between, a smooth transition is built. All are given in units of the diffraction limit.

1.5.1.4. optimization

  • gpu : index of the GPU to be used for the optimization. If you want to use the CPU, just put -1.

  • transform : in case the object is optimized simultaneously with the wavefronts, a transformation can be applied to ensure its positivity. Options are softplus (which uses the softplus function) or none (which uses the identity function).

  • softplus_scale : scale of the softplus function

  • lr_obj : learning rate for the object

  • lr_modes : learning rate for the modes

  • show_object_info : if True, the object information (contrast, minimum and maximum) is shown during the optimization

1.5.1.5. regularization

Different regularizations can be used simultaneously. Current options are smooth (an L2 regularization applied on the spatial derivative of the images) or iuwt (an L1 regularization on the isotropic undecimated wavelet transform).

  • iuwt1iuwt regularization to be used. The parameters are:

    • variable : variable to be regularized. Options are object (which regularizes the object) or modes (which regularizes the modes)

    • lambda : regularization parameter

    • nbands : number of bands for the iuwt

  • smooth1iuwt regularization to be used. The parameters are:

    • variable : variable to be regularized. Options are object (which regularizes the object) or modes (which regularizes the modes)

    • lambda : regularization parameter

1.5.1.6. psf

  • model : model to be used for the PSF. Options are kl (which uses Karhunen-Loeve modes) or zernike (which uses Zernike modes)

  • nmax_modes : number of modes to be used for the PSF

1.5.1.7. initialization

  • object : initialization of the object. Options are contrast (which initializes the object with the contrast) or random (which initializes the object with random values). This has impact only if the ``l2``loss is used.

  • modes_std : standard deviation of the modes at initialization

1.5.1.8. annealing

The modes are annealed during the optimization. This means modes are added slowly while the optimization proceeds. The specific annealing :

  • type : type of annealing. Options are sigmoid (which uses a sigmoid function), linear (which uses a linear function) or none

  • start_pct : percentage of the total number of iterations at which the annealing starts

  • end_pct : percentage of the total number of iterations at which the annealing ends

1.5.2. Spatially variant

telescope:
    diameter: 100.0
    central_obscuration : 0.0

images:
    n_pixel : 256
    wavelength : 8542.0
    pix_size : 0.059
    apodization_border : 6

object1:
    wavelength : 8542.0
    image_filter: tophat
    cutoff : 0.75

object2:
    wavelength : 8542.0
    image_filter: tophat
    cutoff : 0.75

optimization:
    gpu : 0
    transform : softplus
    softplus_scale : 1000.0
    lr_obj : 0.02
    lr_modes : 0.08
    lr_tt : 0.003

regularization:
    smooth1:
        variable: tiptilt
        lambda: 0.01
    smooth2:
        variable: modes
        lambda: 0.01
    iuwt1:
        variable : object
        lambda : 0.00
        nbands : 5

psf:
    model : nmf
    nmax_modes : 150
    filename: '../basis/nmf_8542_n_150_r0_5_30.npz'
    ngrid_modes: 8

initialization:
    object : contrast
    modes_std : 0.0

annealing:
    type: none
    start_pct : 0.0
    end_pct : 0.85

1.5.2.1. telescope

  • diameter: diameter of the telescope in cm

  • central_obscuration : central obscuration of the telescope in cm

1.5.2.2. images

  • n_pixel : size in pixels of the images to be deconvolved. In case the images are mosaicked to take into account the size of the anisoplanatic patch, this keyword gives the size of the patches

  • pix_size : pixel size in arcsec

  • apodization_border : border of the apodization in pixels to avoid edge effects in the Fourier transform used for the convolution

1.5.2.3. objects

Since several objects can be deconvolved simultaneously, one needs to define the parameters for each object. Objects are defined as objectx, where x is an index starting from 1.

  • wavelength : wavelength of the object in angstroms

  • image_filter : Fourier filter to be used for the image. Options are tophat (a simple tophat filter to avoid frequencies above the cutoff) or scharmer (the filter from Lofdahl & Scharmer 1994).

  • cutoff : cutoff frequency indicated in units of the diffraction limit

1.5.2.4. optimization

  • gpu : index of the GPU to be used for the optimization. If you want to use the CPU, just put -1.

  • transform : in case the object is optimized simultaneously with the wavefronts, a transformation can be applied to ensure its positivity. Options are softplus (which uses the softplus function) or none (which uses the identity function).

  • softplus_scale : scale of the softplus function

  • lr_obj : learning rate for the object

  • lr_tt : learning rate for the tip-tilt

  • lr_modes : learning rate for the modes

1.5.2.5. regularization

Different regularizations can be used simultaneously. Current options are smooth (an L2 regularization applied on the spatial derivative of the images) or iuwt (an L1 regularization on the isotropic undecimated wavelet transform).

  • iuwt1iuwt regularization to be used. The parameters are:

    • variable : variable to be regularized. Options are object (which regularizes the object) or modes (which regularizes the modes)

    • lambda : regularization parameter

    • nbands : number of bands for the iuwt

  • smooth1iuwt regularization to be used. The parameters are:

    • variable : variable to be regularized. Options are object (which regularizes the object) or modes (which regularizes the modes)

    • lambda : regularization parameter

1.5.2.6. psf

  • model : model to be used for the PSF. Options are nmf (which uses non-negative matrix factorization) or pca (which uses principal component analysis)

  • nmax_modes : number of modes to be used for the PSF

  • filename : filename of the basis to be used for the PSF

  • ngrid_modes : number of grid points where to infer the PSF modes. They will be interpolated to the full size of the image

1.5.2.7. initialization

  • object : initialization of the object. Options are contrast (which initializes the object with the contrast) or random (which initializes the object with random values). This has impact only if the object is inferred.

  • modes_std : standard deviation of the modes at initialization

1.5.2.8. annealing

The modes are annealed during the optimization. This means modes are added slowly while the optimization proceeds. The specific annealing :

  • type : type of annealing. Options are sigmoid (which uses a sigmoid function), linear (which uses a linear function) or none

  • start_pct : percentage of the total number of iterations at which the annealing starts

  • end_pct : percentage of the total number of iterations at which the annealing ends