Optimization

class CALOptimize(target_obj, opt_params, proj_params, verbose)

Constructor for the CALOptimize class that handles optimization for a given input target object, target_obj.

Parameters

  • target_obj - TargetObj containing the target (output from CALPrepTarget())

  • opt_params - structure of optimization parameters defining the optimization configuration

    • parallel - activates parallel processing if set to 1

    • max_iter - number of iterations used in the iterative optimization

    • learning_rate- step size in gradient descent

    • sigmoid - width of the sigmoid thresholding function, smaller corresponds to higher contrast photoresist

    • threshold - threshold value from 0 to 1, if NaN, the threshold will float to the optimum during optimization

    • Beta - optimization memory/momentum, 0 to 1, higher value gives more weight to previous gradients and can lead to faster convergence

    • Theta - relaxation of positivity constraint, 0 to 1, higher value relaxes positivity constraint more

    • Rho - robustness for dilation and erosion, 0 to 1, higher value enforces higher accuracy in thin region around target contour

    opt_params.x

    Default value

    parallel

    0

    max_iter

    10

    learning_rate

    0.005

    sigmoid

    0.01

    threshold

    NaN

    Beta

    0

    Theta

    0

    Rho

    0
  • proj_params - structure of projection parameters defining the projection configuration

    • angles - vector of angles at which to perform forward and back projection

    • bit8 - activates projection with simulated 8-bit intensity (pixel intensities are binned into 256 bins)

    proj_params.x

    Default value

    angles

    [0:1:179]

    bit8

    0

  • verbose - to show additional details and plots during runtime

Returns
classmethod run(obj)
Parameters

  • None

Returns

  • opt_proj_obj - ProjObj containing the optimized projections

  • opt_recon_obj - ReconObj containing the optimized reconstruction

  • obj - updated instance of CALOptimize containing the error and thresholds (if floating) calculated during optimization.

classmethod getInds(obj)
Parameters

  • None

Returns

  • gel_inds - indices for the voxels/pixels inside the target

  • void_inds - indices for the voxels/pixels outside the target and inside the inscribed reconstruction circle

classmethod evalError(obj, x)
Parameters

  • x - reconstruction on which to evaluate the error

Returns

  • VER - voxel error rate, the error evaluated as \(VER=\frac{W}{N}\) where \(W\) is the number void voxels that receive more dose than the dose of the minimum gel voxel and \(N\) is the total number of gel and void voxels

classmethod sigmoid(x, g)
Parameters

  • x - reconstruction to threshold

  • g - sigmoid width

Returns

  • y - sigmoid thresholded reconstruction