TVL1Denoiser#

class deepinv.models.TVL1Denoiser(verbose=False, tau=0.01, rho=1.99, n_it_max=1000, crit=1e-5, x2=None, u2=None, ths=None)[source]#

Bases: TVDenoiser

Compute the proximal operator of the conjugate TV-L1 regularization term.

This operator projects the input tensor onto the interval [-lambda2, lambda2] by clamping each element independently.

Parameters:

  • u (torch.Tensor) – Input dual variable tensor.

  • lambda2 (float or torch.Tensor) – Clamping threshold defining the projection bounds.

Returns:

Tensor with all values constrained to the interval [-lambda2, lambda2].

Return type:

torch.Tensor