MLI-lab / overparameterized_convolutional_generators

Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators

This repository provides code for reproducing the figures in the paper:

``Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators'', by Reinhard Heckel and Mahdi Soltanolkotabi. Contact: [e-mail]

The paper is available online here.

Organization

• Figure 1: denoising_MSE_curves.ipynb
• Figure 2: denoising_performance_example.ipynb, denoising_bm3d_example.ipynb
• Figure 4,8: noise_vs_img_fitting_different_architectures.ipynb
• Figure 5: linear_least_squares_selective_fitting_warmup.ipynb
• Figure 6: kernels_and_associated_dual_kernels.ipynb
• Figure 7: Jacobian_multi_layer_deep_decoder.ipynb
• Figure 10: image_fitted_faster_than_noise_on_imgnet.ipynb
• Figure 12: Jacobian_inner_product_noisevsimg.ipynb
• Table 1: denoising_imagenet_selected100_paper.ipynb, denoising_bm3d_imagenet_selected100_paper.ipynb

Installation

The code is written in python and relies on pytorch. The following libraries are required:

• python 3
• pytorch
• numpy
• skimage
• matplotlib
• scikit-image
• jupyter

The libraries can be installed via:

conda install jupyter

A small part of the code compares performance to the deep image prior. This part requires downloading the models folder from https://github.com/DmitryUlyanov/deep-image-prior.

Citation

@article{heckel_denoising_2019,
    author    = {Reinhard Heckel and Mahdi Soltanolkotabi},
    title     = {Denoising and Regularization via Exploiting the Structural Bias of Convolutional Generators},
    journal   = {arXiv:1910.14634 [cs.LG]},
    year      = {2019}
}

Licence

All files are provided under the terms of the Apache License, Version 2.0.