Proximal Operators

Proximal Operators#

Sub-package containing regularization routines.

DeepMR provides regularization routines for plug-n-play compressed sensing.

Classical Denoisers#

Classical (i.e., non-trainable) denoisers can be used inside iterative reconstructions as regularizars (aka, PnP reconstruction).

DeepMR expose denoisers both as torch.nn objects (to be chained in DL-based reconstructions) and functional forms (e.g., for simple standalone denoising).

Currently available denoisers are Wavelet (WaveletDenoiser, WaveletDictDenoiser), Local Low Rank (LLRDenoiser) and Total (Generalized) Variation (TVDenoiser, ``TGVDenoiser`).

Both WaveletDenoiser / WaveletDictDenoiser and TVDenoiser / TGVDenoiser are adapted for complex-value inputs from the corresponding DeepInverse implementations.

deepmr.prox.WaveletDenoiser

Orthogonal Wavelet denoising with the \(\ell_1\) norm.

deepmr.prox.WaveletDictDenoiser

Overcomplete Wavelet denoising with the \(\ell_1\) norm.

deepmr.prox.TVDenoiser

Proximal operator of the isotropic Total Variation operator.

deepmr.prox.TGVDenoiser

Proximal operator of (2nd order) Total Generalised Variation operator.

deepmr.prox.LLRDenoiser

Local Low Rank denoising.

deepmr.prox.wavelet_dict_denoise

Apply overcomplete Wavelet denoising with the \(\ell_1\) norm.

deepmr.prox.wavelet_denoise

Apply orthogonal Wavelet denoising with the \(\ell_1\) norm.

deepmr.prox.tv_denoise

Apply isotropic Total Variation denoising.

deepmr.prox.tgv_denoise

Apply Total Generalized Variation denoising.

deepmr.prox.llr_denoise

Apply Local Low Rank denoising.