# Loss

### *class* deepinv.loss.Loss

Bases: [`Module`](https://docs.pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module)

Base class for all loss functions.

Sets a template for the loss functions, whose forward method must follow the input parameters in
[`deepinv.loss.Loss.forward()`](#deepinv.loss.Loss.forward).

#### adapt_model(model, \*\*kwargs)

Some loss functions require the model forward call to be adapted before the forward pass.

* **Parameters:**
  **model** ([*torch.nn.Module*](https://docs.pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module)) – reconstruction model

#### forward(x_net, x, y, physics, model, \*\*kwargs)

Computes the loss.

* **Parameters:**
  * **x_net** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Reconstructed image $\inverse{y}$.
  * **x** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Reference image.
  * **y** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Measurement.
  * **physics** ([*deepinv.physics.Physics*](https://deepinv.org/api/stubs/deepinv.physics.Physics.html.md#deepinv.physics.Physics)) – Forward operator associated with the measurements.
  * **model** ([*torch.nn.Module*](https://docs.pytorch.org/docs/stable/generated/torch.nn.Module.html#torch.nn.Module)) – Reconstruction function.
* **Returns:**
  ([`torch.Tensor`](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) loss, the tensor size might be (1,) or (batch size,).
* **Return type:**
  [torch.Tensor](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)

<a id="sphx-glr-backref-deepinv-loss-loss"></a>

## Examples using `Loss`:

<div class="sphx-glr-thumbnails">
<!-- thumbnail-parent-div-open --><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train various networks using adversarial training for deblurring problems. We demonstrate running training and inference using a conditional GAN (i.e. DeblurGAN), CSGM, AmbientGAN and UAIR implemented in the library, and how to simply train your own GAN by using deepinv.training.AdversarialTrainer. These examples can also be easily extended to train more complicated GANs such as CycleGAN.">![](auto_examples/adversarial-learning/images/thumb/sphx_glr_demo_gan_imaging_thumb.png)

[Imaging inverse problems with adversarial networks](https://deepinv.org/auto_examples/adversarial-learning/demo_gan_imaging.html.md)

  <div class="sphx-glr-thumbnail-title">Imaging inverse problems with adversarial networks</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="In many large-scale imaging problems, the size of the image/volume to reconstruct is very large, making it impossible to train reconstruction networks (in this example, unfolded networks) with a single GPU. The deepinv.distributed framework enables training a model on multiple GPUs, by carefully parallelizing the data fidelity and denoising steps inside the network.">![](auto_examples/distributed/images/thumb/sphx_glr_demo_unrolled_distributed_thumb.png)

[Distributed Training of Unfolded Networks](https://deepinv.org/auto_examples/distributed/demo_unrolled_distributed.html.md)

  <div class="sphx-glr-thumbnail-title">Distributed Training of Unfolded Networks</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows how to perform inference on and fine-tune the Reconstruct Anything Model (RAM) foundation model terris2025reconstruct to solve inverse problems.">![](auto_examples/models/images/thumb/sphx_glr_demo_foundation_model_thumb.png)

[Inference and fine-tune a foundation model](https://deepinv.org/auto_examples/models/demo_foundation_model.html.md)

  <div class="sphx-glr-thumbnail-title">Inference and fine-tune a foundation model</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="Single-image super-resolution (SISR) is the inverse problem of recovering a high-resolution (HR) image x from a low-resolution (LR) observation y = \\downarrow_s(x), where \\downarrow_s denotes downsampling by factor s.">![](auto_examples/models/images/thumb/sphx_glr_demo_super_resolution_thumb.png)

[Super-resolution with SRResNet](https://deepinv.org/auto_examples/models/demo_super_resolution.html.md)

  <div class="sphx-glr-thumbnail-title">Super-resolution with SRResNet</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example provides a very simple quick start introduction to training reconstruction networks with DeepInverse for solving imaging inverse problems.">![](auto_examples/models/images/thumb/sphx_glr_demo_training_thumb.png)

[Training a reconstruction model](https://deepinv.org/auto_examples/models/demo_training.html.md)

  <div class="sphx-glr-thumbnail-title">Training a reconstruction model</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example presents the various datasets, forward physics and models available in DeepInverse for Magnetic Resonance Imaging (MRI) problems:">![](auto_examples/physics/images/thumb/sphx_glr_demo_mri_tour_thumb.png)

[Tour of MRI functionality in DeepInverse](https://deepinv.org/auto_examples/physics/demo_mri_tour.html.md)

  <div class="sphx-glr-thumbnail-title">Tour of MRI functionality in DeepInverse</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="In this example we demonstrate remote sensing inverse problems for multispectral satellite imaging.">![](auto_examples/physics/images/thumb/sphx_glr_demo_remote_sensing_thumb.png)

[Remote sensing with satellite images](https://deepinv.org/auto_examples/physics/demo_remote_sensing.html.md)

  <div class="sphx-glr-thumbnail-title">Remote sensing with satellite images</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="We demonstrate the self-supervised Artifact2Artifact loss for solving an undersampled sequential MRI reconstruction problem without ground truth.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_artifact2artifact_thumb.png)

[Self-supervised MRI reconstruction with Artifact2Artifact](https://deepinv.org/auto_examples/self-supervised-learning/demo_artifact2artifact.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised MRI reconstruction with Artifact2Artifact</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example demonstrates various geometric image transformations implemented in deepinv that can be used in Equivariant Imaging (EI) for self-supervised learning:">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_ei_transforms_thumb.png)

[Image transformations for Equivariant Imaging](https://deepinv.org/auto_examples/self-supervised-learning/demo_ei_transforms.html.md)

  <div class="sphx-glr-thumbnail-title">Image transformations for Equivariant Imaging</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a reconstruction network for an MRI inverse problem on a fully self-supervised way, i.e., using measurement data only.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_equivariant_imaging_thumb.png)

[Self-supervised learning with Equivariant Imaging for MRI.](https://deepinv.org/auto_examples/self-supervised-learning/demo_equivariant_imaging.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised learning with Equivariant Imaging for MRI.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="Equivariant splitting consists in minimizing a self-supervised loss to train a reconstruction model using measurement data only sechaud26Equivariant.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_equivariant_splitting_thumb.png)

[Self-supervised learning with Equivariant Splitting](https://deepinv.org/auto_examples/self-supervised-learning/demo_equivariant_splitting.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised learning with Equivariant Splitting</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="We demonstrate self-supervised (blind) denoising of a low-field MRI scan without ground truth data.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_lowfieldmri_thumb.png)

[Low-field MRI denoising without ground truth](https://deepinv.org/auto_examples/self-supervised-learning/demo_lowfieldmri.html.md)

  <div class="sphx-glr-thumbnail-title">Low-field MRI denoising without ground truth</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a reconstruction network for an inpainting inverse problem on a fully self-supervised way, i.e., using measurement data only.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_multioperator_imaging_thumb.png)

[Self-supervised learning from incomplete measurements of multiple operators.](https://deepinv.org/auto_examples/self-supervised-learning/demo_multioperator_imaging.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised learning from incomplete measurements of multiple operators.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a denoiser network in a fully self-supervised way, i.e., using noisy images only via the Neighbor2Neighbor loss, which exploits the local correlation of natural images.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_n2n_denoising_thumb.png)

[Self-supervised denoising with the Neighbor2Neighbor loss.](https://deepinv.org/auto_examples/self-supervised-learning/demo_n2n_denoising.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised denoising with the Neighbor2Neighbor loss.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a denoiser network in a fully self-supervised way, using noisy images only via the Generalized Recorrupted2Recorrupted (GR2R) loss monroy2025generalized, which exploits knowledge about the noise distribution. You can change the noise distribution by selecting from predefined noise models such as Gaussian, Poisson, and Gamma noise.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_r2r_denoising_thumb.png)

[Self-supervised denoising with the Generalized R2R loss.](https://deepinv.org/auto_examples/self-supervised-learning/demo_r2r_denoising.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised denoising with the Generalized R2R loss.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="We demonstrate scan-specific self-supervised learning, that is, learning to reconstruct MRI scans from a single accelerated sample without ground truth.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_scan_specific_thumb.png)

[Scan-specific zero-shot SSDU for MRI](https://deepinv.org/auto_examples/self-supervised-learning/demo_scan_specific.html.md)

  <div class="sphx-glr-thumbnail-title">Scan-specific zero-shot SSDU for MRI</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="We demonstrate self-supervised learning with measurement splitting, to train a denoiser network on the MNIST dataset. The physics here is noisy computed tomography, as is the case in Noise2Inverse hendriksen2020noise2inverse. Note this example can also be easily applied to undersampled multicoil MRI as is the case in SSDU yaman2020self.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_splitting_loss_thumb.png)

[Self-supervised learning with measurement splitting](https://deepinv.org/auto_examples/self-supervised-learning/demo_splitting_loss.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised learning with measurement splitting</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a denoiser network in a fully self-supervised way, i.e., using noisy images only via the SURE loss, which exploits knowledge about the noise distribution.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_sure_denoising_thumb.png)

[Self-supervised denoising with the SURE loss.](https://deepinv.org/auto_examples/self-supervised-learning/demo_sure_denoising.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised denoising with the SURE loss.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows you how to train a denoiser network in a fully self-supervised way, i.e., using noisy images with unknown noise level only via the UNSURE loss, which is introduced by tachella2024unsure.">![](auto_examples/self-supervised-learning/images/thumb/sphx_glr_demo_unsure_thumb.png)

[Self-supervised denoising with the UNSURE loss.](https://deepinv.org/auto_examples/self-supervised-learning/demo_unsure.html.md)

  <div class="sphx-glr-thumbnail-title">Self-supervised denoising with the UNSURE loss.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="We demonstrate the use of our deepinv.transform module for use in solving imaging problems. These can be used for:">![](auto_examples/transforms-equivariance/images/thumb/sphx_glr_demo_transforms_thumb.png)

[Image transforms for equivariance & augmentations](https://deepinv.org/auto_examples/transforms-equivariance/demo_transforms.html.md)

  <div class="sphx-glr-thumbnail-title">Image transforms for equivariance & augmentations</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This a toy example to show you how to use DEQ to solve a deblurring problem. Note that this is a small dataset for training. For optimal results, use a larger dataset.">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_DEQ_thumb.png)

[Deep Equilibrium (DEQ) algorithms for image deblurring](https://deepinv.org/auto_examples/unfolded/demo_DEQ.html.md)

  <div class="sphx-glr-thumbnail-title">Deep Equilibrium (DEQ) algorithms for image deblurring</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows how to implement the LISTA algorithm gregor2010learning, for a compressed sensing problem. In a nutshell, LISTA is an unfolded proximal gradient algorithm involving a soft-thresholding proximal operator with learnable thresholding parameters.">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_LISTA_thumb.png)

[Learned Iterative Soft-Thresholding Algorithm (LISTA) for compressed sensing](https://deepinv.org/auto_examples/unfolded/demo_LISTA.html.md)

  <div class="sphx-glr-thumbnail-title">Learned Iterative Soft-Thresholding Algorithm (LISTA) for compressed sensing</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This example shows how to implement a learned unrolled proximal gradient descent algorithm with a custom prior function. The custom prior in use is The algorithm is trained on a dataset of compressed sensing measurements of MNIST images.">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_custom_prior_unfolded_thumb.png)

[Learned iterative custom prior](https://deepinv.org/auto_examples/unfolded/demo_custom_prior_unfolded.html.md)

  <div class="sphx-glr-thumbnail-title">Learned iterative custom prior</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="where both the data fidelity and the prior are learned modules, distinct for each iterations.">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_learned_primal_dual_thumb.png)

[Learned Primal-Dual algorithm for CT scan.](https://deepinv.org/auto_examples/unfolded/demo_learned_primal_dual.html.md)

  <div class="sphx-glr-thumbnail-title">Learned Primal-Dual algorithm for CT scan.</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="Image inpainting consists in solving y = Ax where A is a mask operator. This problem can be reformulated as the following minimization problem:">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_unfolded_constrained_LISTA_thumb.png)

[Unfolded Chambolle-Pock for constrained image inpainting](https://deepinv.org/auto_examples/unfolded/demo_unfolded_constrained_LISTA.html.md)

  <div class="sphx-glr-thumbnail-title">Unfolded Chambolle-Pock for constrained image inpainting</div>
</div><div class="sphx-glr-thumbcontainer" tooltip="This is a simple example to show how to use vanilla unfolded Plug-and-Play. The DnCNN denoiser and the algorithm parameters (stepsize, regularization parameters) are trained jointly. For simplicity, we show how to train the algorithm on a  small dataset. For optimal results, use a larger dataset.">![](auto_examples/unfolded/images/thumb/sphx_glr_demo_vanilla_unfolded_thumb.png)

[Vanilla Unfolded algorithm for super-resolution](https://deepinv.org/auto_examples/unfolded/demo_vanilla_unfolded.html.md)

  <div class="sphx-glr-thumbnail-title">Vanilla Unfolded algorithm for super-resolution</div>
</div>
<!-- thumbnail-parent-div-close --></div>