Skip to content

Losses

Source

Losses module.

denoisplit_loss(model_outputs, targets, config, gaussian_likelihood=None, noise_model_likelihood=None)

Loss function for DenoiSplit.

Parameters:

Name Type Description Default
model_outputs tuple[Tensor, dict[str, Any]]

Tuple containing the model predictions (shape is (B, target_ch, [Z], Y, X)) and the top-down layer data (e.g., sampled latents, KL-loss values, etc.).

 required
targets Tensor

The target image used to compute the reconstruction loss. Shape is (B, target_ch, [Z], Y, X).

 required
config LVAELossConfig

The config for loss function containing all loss hyperparameters.

 required
gaussian_likelihood GaussianLikelihood

The Gaussian likelihood object.

 None
noise_model_likelihood NoiseModelLikelihood

The noise model likelihood object.

 None

Returns:

Name Type Description
output Optional[dict[str, Tensor]]

A dictionary containing the overall loss ["loss"], the reconstruction loss ["reconstruction_loss"], and the KL divergence loss ["kl_loss"].

denoisplit_musplit_loss(model_outputs, targets, config, gaussian_likelihood, noise_model_likelihood)

Loss function for DenoiSplit.

Parameters:

Name Type Description Default
model_outputs tuple[Tensor, dict[str, Any]]

Tuple containing the model predictions (shape is (B, target_ch, [Z], Y, X)) and the top-down layer data (e.g., sampled latents, KL-loss values, etc.).

 required
targets Tensor

The target image used to compute the reconstruction loss. Shape is (B, target_ch, [Z], Y, X).

 required
config LVAELossConfig

The config for loss function containing all loss hyperparameters.

 required
gaussian_likelihood GaussianLikelihood

The Gaussian likelihood object.

 required
noise_model_likelihood NoiseModelLikelihood

The noise model likelihood object.

 required

Returns:

Name Type Description
output Optional[dict[str, Tensor]]

A dictionary containing the overall loss ["loss"], the reconstruction loss ["reconstruction_loss"], and the KL divergence loss ["kl_loss"].

hdn_loss(model_outputs, targets, config, gaussian_likelihood, noise_model_likelihood)

Loss function for HDN.

Parameters:

Name Type Description Default
model_outputs tuple[Tensor, dict[str, Any]]

Tuple containing the model predictions (shape is (B, target_ch, [Z], Y, X)) and the top-down layer data (e.g., sampled latents, KL-loss values, etc.).

 required
targets Tensor

The target image used to compute the reconstruction loss. In this case we use the input patch itself as target. Shape is (B, target_ch, [Z], Y, X).

 required
config LVAELossConfig

The config for loss function containing all loss hyperparameters.

 required
gaussian_likelihood GaussianLikelihood

The Gaussian likelihood object.

 required
noise_model_likelihood NoiseModelLikelihood

The noise model likelihood object.

 required

Returns:

Name Type Description
output Optional[dict[str, Tensor]]

A dictionary containing the overall loss ["loss"], the reconstruction loss ["reconstruction_loss"], and the KL divergence loss ["kl_loss"].

mae_loss(samples, labels, *args)

N2N Loss function described in to J Lehtinen et al 2018.

Parameters:

Name Type Description Default
samples Tensor

Raw patches.

 required
labels Tensor

Different subset of noisy patches.

 required
*args Any

Additional arguments.

 ()

Returns:

Type Description
Tensor

Loss value.

mse_loss(source, target, *args)

Mean squared error loss.

Parameters:

Name Type Description Default
source Tensor

Source patches.

 required
target Tensor

Target patches.

 required
*args Any

Additional arguments.

 ()

Returns:

Type Description
Tensor

Loss value.

musplit_loss(model_outputs, targets, config, gaussian_likelihood, noise_model_likelihood=None)

Loss function for muSplit.

Parameters:

Name Type Description Default
model_outputs tuple[Tensor, dict[str, Any]]

Tuple containing the model predictions (shape is (B, target_ch, [Z], Y, X)) and the top-down layer data (e.g., sampled latents, KL-loss values, etc.).

 required
targets Tensor

The target image used to compute the reconstruction loss. Shape is (B, target_ch, [Z], Y, X).

 required
config LVAELossConfig

The config for loss function (e.g., KL hyperparameters, likelihood module, noise model, etc.).

 required
gaussian_likelihood GaussianLikelihood

The Gaussian likelihood object.

 required
noise_model_likelihood Optional[NoiseModelLikelihood]

The noise model likelihood object. Not used here.

 None

Returns:

Name Type Description
output Optional[dict[str, Tensor]]

A dictionary containing the overall loss ["loss"], the reconstruction loss ["reconstruction_loss"], and the KL divergence loss ["kl_loss"].

n2v_loss(manipulated_batch, original_batch, masks, *args)

N2V Loss function described in A Krull et al 2018.

Parameters:

Name Type Description Default
manipulated_batch Tensor

Batch after manipulation function applied.

 required
original_batch Tensor

Original images.

 required
masks Tensor

Coordinates of changed pixels.

 required
*args Any

Additional arguments.

 ()

Returns:

Type Description
Tensor

Loss value.