ConvLista

class deepinv.models.ConvLista(*, in_channels, out_channels, kernel_size=3, num_filters=512, stride=2, num_iter=10, threshold=1e-2)

Bases: Module

Convolutional LISTA network.

The architecture was introduced by Simon and Elad[1], and it is well suited as a backbone for Poisson2Sparse (see deepinv.models.Poisson2Sparse).

Note

The decoder expects images with a dynamic range normalized between zero and one.

Parameters:

• in_channels (int) – Number of channels in the input image.

• out_channels (int) – Number of channels in the output image.

• kernel_size (int) – Size of the convolutional kernels (default: 3).

• num_filters (int) – Number of filters in the convolutional layers (default: 512).

• stride (int) – Stride of the convolutional layers (default: 2).

• num_iter (int) – Number of iterations of the LISTA algorithm (default: 10).

• threshold (float) – Initial value for the learned soft-thresholding (default: 1e-2).

---

References:

[1]

Dror Simon and Michael Elad. Rethinking the csc model for natural images. Advances in Neural Information Processing Systems, 2019.

Examples using ConvLista:

Poisson denoising using Poisson2Sparse

Poisson denoising using Poisson2Sparse