Learning to Recorrupt: Noise Distribution Agnostic Self-Supervised Image Denoising

Brayan Monroy, Jorge Bacca, Julian Tachella

---

In this paper, we present Learning to Recorrupt (L2R), a noise distribution-agnostic denoising technique that eliminates the need for knowledge of the noise distribution. Our method introduces a learnable monotonic neural network that learns the recorruption process through a min-max saddle-point objective. The proposed method achieves state-of-the-art performance across unconventional and heavy-tailed noise distributions, such as log-gamma, Laplace, and spatially correlated noise, as well as signal-dependent noise models such as Poisson-Gaussian noise.

Method

We introduce L2R, a loss for unsupervised image denoising with unorganized noisy images when the observation model $y \sim p(y \mid x)$ is unknown. Specifically, we define a trainable recorruption mechanism:

$$ y_1 = y + \tau h(\omega) $$

where $h$ is a monotonic neural network that maps additive white Gaussian noise (AWGN) $\omega$ to the desired recorruption noise. The L2R objective is then computed adversarially as:

$$ \min_{f} \max_{h} \Vert f(y_1) - y \Vert^2_2 + \frac{2}{\tau} f(y_1)^\top h(\omega). $$

How to cite

If this code is useful for your research and you use it in an academic work, please consider citing this paper as

@article{monroy2026learning,
  title={Learning to Recorrupt: Noise Distribution Agnostic Self-Supervised Image Denoising},
  author={Monroy, Brayan and Bacca, Jorge and Tachella, Juli{\'a}n},
  journal={arXiv preprint arXiv:2603.25869},
  year={2026}
}