Poisson2Sparse

class deepinv.models.Poisson2Sparse(backbone=None, *, lr=1e-4, weight_n2n=2.0, weight_l1_regularization=1e-5, num_iter=200, verbose=False)

Bases: Denoiser

Poisson2Sparse model for Poisson denoising.

This method, introduced by Ta et al.[1], reconstructs an image corrupted by Poisson noise by learning a sparse non-linear dictionary parametrized by a neural network using a combination of Neighbor2Neighbor Huang et al.[2], of the negative log Poisson likelihood, of the \(\ell^1\) pixel distance and of a sparsity-inducing \(\ell^1\) regularization function on the weights.

Note

This method does not use knowledge of the physics model and assumes a Poisson degradation model internally. Therefore, the physics object can be omitted when calling the model and specifying it will have no effect.

Note

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

Parameters:

• backbone (torch.nn.Module, None) – Neural network used as a non-linear dictionary. If None, a default deepinv.models.ConvLista model is used.

• lr (float) – Learning rate of the AdamW optimizer (default: 1e-4).

• weight_n2n (float) – Weight of the Neighbor2Neighbor loss term (default: 2.0).

• weight_l1_regularization (float) – Weight of the sparsity-inducing \(\ell^1\) regularization on the weights (default: 1e-5).

• num_iter (int) – Number of optimization iterations (default: 200).

• verbose (bool) – If True, print progress (default: False).

---

References:

[1]

Calvin-Khang Ta, Abhishek Aich, Akash Gupta, and Amit K Roy-Chowdhury. Poisson2sparse: self-supervised poisson denoising from a single image. In International Conference on Medical Image Computing and Computer-Assisted Intervention, 557–567. Springer, 2022.

[2]

Tao Huang, Songjiang Li, Xu Jia, Huchuan Lu, and Jianzhuang Liu. Neighbor2neighbor: self-supervised denoising from single noisy images. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 14781–14790. 2021.

Examples using Poisson2Sparse:

Poisson denoising using Poisson2Sparse

Poisson denoising using Poisson2Sparse