A simple way to make neural networks robust against diverse image corruptions

Evgenia Rusak, Lukas Schott, Roland S. Zimmermann, Julian Bitterwolf, Oliver Bringmann +2 more
1/16/2020
cs.CVcs.LGstat.ML

Official code: hendrycks/robustness (1,162 stars, Apache-2.0, PyTorch, last commit 2022-08-24). 1 more implementation below.

Abstract

The human visual system is remarkably robust against a wide range of naturally occurring variations and corruptions like rain or snow. In contrast, the performance of modern image recognition models strongly degrades when evaluated on previously unseen corruptions. Here, we demonstrate that a simple but properly tuned training with additive Gaussian and Speckle noise generalizes surprisingly well to unseen corruptions, easily reaching the previous state of the art on the corruption benchmark ImageNet-C (with ResNet50) and on MNIST-C. We build on top of these strong baseline results and show that an adversarial training of the recognition model against uncorrelated worst-case noise distributions leads to an additional increase in performance. This regularization can be combined with previously proposed defense methods for further improvement.

View on arXivView PDF

Code Implementations(2)

hendrycks/robustnessOfficial100%
PyTorchApache-2.0

Corruption and Perturbation Robustness (ICLR 2019)

1,162149Shell, PythonApr 6, 20184 years agoApache-2.0
computer-visionconvolutional-neural-networksdeep-learningdomain-generalizationimagenet+4 more
bethgelab/game-of-noiseOfficial100%
MIT

Trained model weights, training and evaluation code from the paper "A simple way to make neural networks robust against diverse image corruptions"

6212Shell, PythonJan 24, 20203 years agoMIT

Similar Papers

Benchmarking Neural Network Robustness to Common Corruptions and Perturbations

88% match1 repo1.2k2019

Bridging Optimal Transport and Jacobian Regularization by Optimal Trajectory for Enhanced Adversarial Defense

87% match2023

Certified Defenses against Adversarial Examples

87% match2018

Universal Robustness via Median Randomized Smoothing for Real-World Super-Resolution

87% match2024

Assessing Visually-Continuous Corruption Robustness of Neural Networks Relative to Human Performance

86% match2024

A Survey on the Robustness of Computer Vision Models against Common Corruptions

86% match1 repo42023

Cite this paper

@article{rusak2020simple,
  title  = {A simple way to make neural networks robust against diverse image corruptions},
  author = {Evgenia Rusak and Lukas Schott and Roland S. Zimmermann and Julian Bitterwolf and Oliver Bringmann and Matthias Bethge and Wieland Brendel},
  year   = {2020},
  eprint = {2001.06057},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/2001.06057}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2001.06057.svg)](https://codeofpaper.com/paper/2001.06057)

Discussion