Transmittance

class colibri.regularizers.Transmittance(parameter=10, t=0.8)

Bases: Module

Transmittance Regularization for Coded Apeuture Design.

Code adapted from Bacca, Jorge, Tatiana Gelvez-Barrera, and Henry Arguello. “Deep coded aperture design: An end-to-end approach for computational imaging tasks.” IEEE Transactions on Computational Imaging 7 (2021): 1148-1160.

The regularizer computes:

\[\begin{equation} R(\learnedOptics) = \mu \left(\sum_{i=1}^{n}\frac{\learnedOptics_i}{n}-t\right)^2 \end{equation}\]

where \(\mu\) is the regularization parameter and \(t\) is the target transmittance value.

Parameters:

• parameter (float) – Regularization parameter.

• t (float) – Target transmittance value.

forward(x)

Compute transmittance regularization term.

Parameters:

x (torch.Tensor) – Input tensor (layer’s weight).

Returns:

Transmittance regularization term.

Return type:

torch.Tensor

Examples using Transmittance:

Demo Colibri.

Demo Colibri.