# RAM ### *class* deepinv.models.RAM(in_channels=(1, 2, 3), device=None, pretrained=True) Bases: [`Reconstructor`](https://deepinv.org/api/stubs/deepinv.models.Reconstructor.html.md#deepinv.models.Reconstructor), [`Denoiser`](https://deepinv.org/api/stubs/deepinv.models.Denoiser.html.md#deepinv.models.Denoiser) Reconstruct Anything Model (RAM) foundation model. RAM Terris *et al.*[1](#footcite-terris2025reconstruct) is a convolutional neural network model that has been trained to work on a large variety of linear image reconstruction tasks and datasets (deblurring, inpainting, denoising, tomography, MRI, etc.). See [Inference and fine-tune a foundation model](https://deepinv.org/auto_examples/models/demo_foundation_model.html.md#sphx-glr-auto-examples-models-demo-foundation-model-py) for examples on the performance of RAM and how to fine-tune the foundation model on a specific problem and dataset. The model works both as a reconstructor or denoiser: * Reconstructor: RAM takes a [physics operator](https://deepinv.org/user_guide/physics/physics.html.md#physics) `model(y, physics)` with an optional noise model defined in the physics * Denoiser: RAM takes optional Gaussian and/or Poisson noise levels (optionally set to 0) `model(y, sigma=sigma, gamma=gamma)` #### NOTE The physics operator should be normalized (i.e. have unit norm) for best results. Use [`physics.compute_norm()`](https://deepinv.org/api/stubs/deepinv.physics.LinearPhysics.html.md#deepinv.physics.LinearPhysics.compute_norm) to check this. #### TIP This model can handle non-uniform `sigma` and `gain` maps, which can be of size `(batch_size, 1, height, width)`. * **Parameters:** * **in_channels** (*Sequence*) – Number of input channels. If a list is provided, the model will have separate heads for each channel. * **device** ([*str*](https://docs.python.org/3.12/library/stdtypes.html#str)) – Device to which the model should be moved. If None, the model will be created on the default device. * **pretrained** ([*bool*](https://docs.python.org/3.12/library/functions.html#bool) *,* [*str*](https://docs.python.org/3.12/library/stdtypes.html#str)) – If `True`, the model will be initialized with pretrained weights. If `str`, load from file.
* **Examples:** ```pycon >>> import deepinv as dinv >>> x = dinv.utils.load_example("butterfly.png") >>> physics = dinv.physics.Downsampling(filter="bicubic", noise_model=dinv.physics.GaussianNoise(0.01)) >>> y = physics(x) >>> model = dinv.models.RAM() >>> x_hat = model(y, physics) # run model >>> dinv.metric.PSNR()(x_hat, x) > 29.75 tensor([True]) ```
* **References:** * **[1]** Matthieu Terris, Samuel Hurault, Maxime Song, and Julián Tachella. Reconstruct anything model: a lightweight foundation model for computational imaging. *arXiv preprint arXiv:2503.08915*, 2025. #### base_conditioning(x, sigma, gain) Stacks the sigma and gain value as additional channel dimensions to the input tensor. * **Parameters:** * **x** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Input tensor * **sigma** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Gaussian noise level or noise level map * **gain** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Poisson noise gain or Poisson noise map * **Return torch.Tensor:** Input tensor with additional channels for sigma and gain * **Return type:** [*Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor) #### constant2map(value, x) Converts a constant value to a map of the same size as the input tensor x. * **Parameters:** * **value** ([*float*](https://docs.python.org/3.12/library/functions.html#float)) – constant value * **x** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – input tensor * **Return torch.Tensor:** a tensor of size (B, 1, W, H) containing constant maps of shapes (W, H) for each value in the batch. * **Return type:** [*Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor) #### forward(y, physics=None, sigma=None, gain=None, img_size=None) Reconstructs a signal estimate from measurements y #### NOTE The noise levels `sigma` and `gain` can be kept as `None` if a noise model with [`GaussianNoise`](https://deepinv.org/api/stubs/deepinv.physics.GaussianNoise.html.md#deepinv.physics.GaussianNoise), [`PoissonNoise`](https://deepinv.org/api/stubs/deepinv.physics.PoissonNoise.html.md#deepinv.physics.PoissonNoise) or [`PoissonGaussianNoise`](https://deepinv.org/api/stubs/deepinv.physics.PoissonGaussianNoise.html.md#deepinv.physics.PoissonGaussianNoise) is specified in the physics. If both are provided, the `sigma` and `gain` values provided to the model will be used. * **Parameters:** * **y** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – measurements * **physics** ([*deepinv.physics.Physics*](https://deepinv.org/api/stubs/deepinv.physics.Physics.html.md#deepinv.physics.Physics)) – forward operator * **sigma** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Gaussian noise level or noise level map * **gain** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Poisson noise gain or Poisson noise map * **img_size** ([*tuple*](https://docs.python.org/3.12/library/stdtypes.html#tuple)) – (optional) size of the image to reconstruct. If None, will be inferred automatically from the physics. * **Returns:** torch.Tensor: reconstructed signal estimate * **Return type:** [torch.Tensor](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor) #### forward_unet(x0, sigma=None, gain=None, physics=None, y=None) Forward pass of the UNet model. * **Parameters:** * **x0** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – init image * **sigma** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Gaussian noise level or noise level map * **gain** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Poisson noise gain or Poisson noise map * **physics** ([*deepinv.physics.Physics*](https://deepinv.org/api/stubs/deepinv.physics.Physics.html.md#deepinv.physics.Physics)) – physics measurement operator * **y** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – measurements #### get_pad(img_size) Get padding amount for model input. * **Parameters:** **img_size** ([*tuple*](https://docs.python.org/3.12/library/stdtypes.html#tuple)) – model input image shape. * **Return tuple[int, int, int]:** padding amounts for channel dim and spatial dims. * **Return type:** [tuple](https://docs.python.org/3.12/library/stdtypes.html#tuple)[[int](https://docs.python.org/3.12/library/functions.html#int), [int](https://docs.python.org/3.12/library/functions.html#int), [int](https://docs.python.org/3.12/library/functions.html#int)] #### obtain_sigma_gain(physics, sigma, gain, rescale_val, device='cpu') Defines the sigma and gain values to be used in the model. If a noise model is specified in the physics, the sigma and gain values will be taken from the noise model. Else, the sigma and gain values will be set to the thresholds defined in the model (if not provided). * **Parameters:** * **physics** ([*deepinv.physics.LinearPhysics*](https://deepinv.org/api/stubs/deepinv.physics.LinearPhysics.html.md#deepinv.physics.LinearPhysics)) – Physics operator * **sigma** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Gaussian noise level. If None, will be set to the threshold. * **gain** ([*float*](https://docs.python.org/3.12/library/functions.html#float) *,* [*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Poisson noise level. If None, will be set to the threshold. * **rescale_val** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Rescale value to apply to the sigma and gain values. * **device** ([*str*](https://docs.python.org/3.12/library/stdtypes.html#str) *,* [*torch.device*](https://docs.pytorch.org/docs/stable/tensor_attributes.html#torch.device)) – Device to which the sigma and gain values should be moved. #### realign_input(x, physics, y, sigma) Realign the input x based on the measurements y and the physics model. Applies the proximity operator of the L2 norm with respect to the physics model. * **Parameters:** * **x** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Input tensor * **physics** ([*deepinv.physics.Physics*](https://deepinv.org/api/stubs/deepinv.physics.Physics.html.md#deepinv.physics.Physics)) – Physics model * **y** ([*torch.Tensor*](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor)) – Measurements * **Return torch.Tensor:** Realigned input tensor * **Return type:** [torch.Tensor](https://docs.pytorch.org/docs/stable/tensors.html#torch.Tensor) ## Examples using `RAM`:
![](auto_examples/basics/images/thumb/sphx_glr_demo_custom_dataset_thumb.png) [Bring your own dataset](https://deepinv.org/auto_examples/basics/demo_custom_dataset.html.md)
Bring your own dataset
![](auto_examples/basics/images/thumb/sphx_glr_demo_pretrained_model_thumb.png) [Use a pretrained model](https://deepinv.org/auto_examples/basics/demo_pretrained_model.html.md)
Use a pretrained model
![](auto_examples/basics/images/thumb/sphx_glr_demo_quickstart_thumb.png) [5 minute quickstart tutorial](https://deepinv.org/auto_examples/basics/demo_quickstart.html.md)
5 minute quickstart tutorial
![](auto_examples/blind-inverse-problems/images/thumb/sphx_glr_demo_blind_deblurring_thumb.png) [Blind deblurring with kernel estimation network](https://deepinv.org/auto_examples/blind-inverse-problems/demo_blind_deblurring.html.md)
Blind deblurring with kernel estimation network
![](auto_examples/external-libraries/images/thumb/sphx_glr_demo_astra_2detect_thumb.png) [Reconstruct real CT sinograms with the 2DeteCT benchmark](https://deepinv.org/auto_examples/external-libraries/demo_astra_2detect.html.md)
Reconstruct real CT sinograms with the 2DeteCT benchmark
![](auto_examples/external-libraries/images/thumb/sphx_glr_demo_connect_spyrit_thumb.png) [Single-pixel imaging with Spyrit](https://deepinv.org/auto_examples/external-libraries/demo_connect_spyrit.html.md)
Single-pixel imaging with Spyrit
![](auto_examples/external-libraries/images/thumb/sphx_glr_demo_microscopy_denoising_thumb.png) [Low-intensity STED fluorescence microscopy denoising](https://deepinv.org/auto_examples/external-libraries/demo_microscopy_denoising.html.md)
Low-intensity STED fluorescence microscopy denoising
![](auto_examples/metrics/images/thumb/sphx_glr_demo_test_time_tuning_thumb.png) [Blind inverse problems with no reference metrics](https://deepinv.org/auto_examples/metrics/demo_test_time_tuning.html.md)
Blind inverse problems with no reference metrics
![](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)
Inference and fine-tune a foundation model
![](auto_examples/models/images/thumb/sphx_glr_demo_mri_pretrained_thumb.png) [Reconstruct undersampled k-space for cardiac and brain MRI](https://deepinv.org/auto_examples/models/demo_mri_pretrained.html.md)
Reconstruct undersampled k-space for cardiac and brain MRI
![](auto_examples/models/images/thumb/sphx_glr_demo_prospective_mri_thumb.png) [Reconstruct prospectively-undersampled raw multicoil MRI](https://deepinv.org/auto_examples/models/demo_prospective_mri.html.md)
Reconstruct prospectively-undersampled raw multicoil MRI
![](auto_examples/optimization/images/thumb/sphx_glr_demo_multispectral_demosaicing_thumb.png) [Multispectral demosaicing from raw sensor data](https://deepinv.org/auto_examples/optimization/demo_multispectral_demosaicing.html.md)
Multispectral demosaicing from raw sensor data
![](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)
Self-supervised learning with Equivariant Splitting
![](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)
Low-field MRI denoising 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)
Scan-specific zero-shot SSDU for MRI
![](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)
Image transforms for equivariance & augmentations