DnCNN#

class deepinv.models.DnCNN(in_channels=3, out_channels=3, depth=20, bias=True, nf=64, pretrained='download', pretrained_2d_isotropic=False, device='cpu', dim=2)[source]#

Bases: Denoiser

DnCNN convolutional denoiser.

The architecture was introduced by Zhang et al.[1] and is composed of a series of convolutional layers with ReLU activation functions. The number of layers can be specified by the user. Unlike the original paper, this implementation does not include batch normalization layers.

The network can be initialized with pretrained weights, which can be downloaded from an online repository. The pretrained weights are trained with the default parameters of the network, i.e. 20 layers, 64 channels and biases.

Parameters:

  • in_channels (int) – input image channels

  • out_channels (int) – output image channels

  • depth (int) – number of convolutional layers

  • bias (bool) – use bias in the convolutional layers

  • nf (int) – number of channels per convolutional layer

  • pretrained (str, None) – use a pretrained network. If pretrained=None, the weights will be initialized at random using Pytorch’s default initialization. If pretrained='download', the weights will be downloaded from an online repository (only available for architecture with depth 20, 64 channels and biases). It is possible to download weights trained via the regularization method in Pesquet et al.[2], using pretrained='download_lipschitz'. When building a 3D network, it is possible to initialize with 2D pretrained weights by using pretrained='download_2d' or pretrained='download_lipschitz_2d', which provides a good starting point for fine-tuning. Finally, pretrained can also be set as a path to the user’s own pretrained weights. See pretrained-weights for more details.

  • pretrained_2d_isotropic (bool) – when loading 2D pretrained weights into a 3D network, whether to initialize the 3D kernels isotropically. By default the weights are loaded axially, i.e., by initializing the central slice of the 3D kernels with the 2D weights.

  • device (torch.device, str) – Device to put the model on.

  • dim (str, int) – Whether to build 2D or 3D network (if str, can be “2”, “2d”, “3D”, etc.)

References:

forward(x, sigma=None)[source]#

Run the denoiser on noisy image. The noise level is not used in this denoiser.

Parameters:

Note

The argument sigma is included for compatibility with the base class Denoiser but is not used in this model.

Examples using DnCNN:#

Benchmarking pretrained denoisers

Benchmarking pretrained denoisers

Tour of MRI functionality in DeepInverse

Tour of MRI functionality in DeepInverse

Pattern Ordering in a Compressive Single Pixel Camera

Pattern Ordering in a Compressive Single Pixel Camera

PnP with custom optimization algorithm (Primal-Dual Condat-Vu)

PnP with custom optimization algorithm (Primal-Dual Condat-Vu)

Plug-and-Play algorithm with Mirror Descent for Poisson noise inverse problems.

Plug-and-Play algorithm with Mirror Descent for Poisson noise inverse problems.

Vanilla PnP for computed tomography (CT).

Vanilla PnP for computed tomography (CT).

Uncertainty quantification with PnP-ULA.

Uncertainty quantification with PnP-ULA.

Deep Equilibrium (DEQ) algorithms for image deblurring

Deep Equilibrium (DEQ) algorithms for image deblurring

Reducing the memory and computational complexity of unfolded network training

Reducing the memory and computational complexity of unfolded network training

Vanilla Unfolded algorithm for super-resolution

Vanilla Unfolded algorithm for super-resolution