StackedPhysicsLoss#

class deepinv.loss.StackedPhysicsLoss(losses)[source]#

Bases: Loss

Loss function for stacked physics operators.

Adapted to deepinv.physics.StackedPhysics physics composed of multiple physics operators.

Parameters:

losses (list[deepinv.loss.Loss]) – list of loss functions for each physics operator.

Examples:

Gaussian and Poisson losses function for a stacked physics operator:

>>> import torch
>>> import deepinv as dinv
>>> # define two observations, one with Gaussian noise and one with Poisson noise
>>> physics1 = dinv.physics.Denoising(dinv.physics.GaussianNoise(.1))
>>> physics2 = dinv.physics.Denoising(dinv.physics.PoissonNoise(.1))
>>> physics = dinv.physics.StackedLinearPhysics([physics1, physics2])
>>> loss1 = dinv.loss.SureGaussianLoss(.1)
>>> loss2 = dinv.loss.SurePoissonLoss(.1)
>>> loss = dinv.loss.StackedPhysicsLoss([loss1, loss2])
>>> x = torch.ones(1, 1, 5, 5) # image
>>> y = physics(x) # noisy measurements
>>> # define a denoiser model
>>> model = dinv.models.ArtifactRemoval(dinv.models.MedianFilter(3))
>>> x_net = model(y, physics)
>>> l = loss(x_net, x, y, physics, model)
forward(x_net: Tensor, x: Tensor, y: TensorList, physics: StackedPhysics, model: Module, **kwargs) → Tensor[source]#

Computes the loss as

\[\sum_i \mathcal{L}_i(x, y_i, \inverse{y}, \physics_i, \model),\]

where \(i\) is the index of the physics operator in the stacked physics.

Parameters:

Examples using StackedPhysicsLoss:#

Remote sensing with satellite images

Remote sensing with satellite images