arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2001.06057cs.CVcs.LGstat.ML
收藏

一种使神经网络对不同图像损坏具有鲁棒性的简单方法

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

University of Tübingen(蒂宾根大学);International Max Planck Research School for Intelligent Systems(国际马克斯·普朗克智能系统研究学院)

Evgenia Rusak, Lukas Schott, Roland S. Zimmermann, Julian Bitterwolf, Oliver Bringmann, Matthias Bethge, Wieland Brendel

摘要:

人类的视觉系统对各种自然发生的变化和破坏(如雨或雪)非常强大。相比之下,现代图像识别模型的性能大大降低时,以前看不见的腐败进行评估。在这里,我们证明了一个简单但经过适当调整的加性高斯和斑点噪声训练可以令人惊讶地很好地推广到看不见的腐败,很容易达到腐败基准ImageNet-C(使用ResNet 50)和MNIST-C的最新技术水平。我们建立在这些强大的基线结果之上,并表明针对不相关的最坏情况噪声分布的识别模型的对抗性训练会导致性能的额外提高。这种正则化可以与以前提出的防御方法相结合,以进一步改进。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑