AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty

Modern deep neural networks can achieve high accuracy when the training distribution and test distribution are identically distributed, but this assumption is frequently violated in practice. When the train and test distributions are mismatched, accuracy can plummet. Currently there are few techniques that improve robustness to unforeseen data shifts encountered during deployment. In this work, we propose a technique to improve the robustness and uncertainty estimates of image classifiers. We propose AugMix, a data processing technique that is simple to implement, adds limited computational overhead, and helps models withstand unforeseen corruptions. AugMix significantly improves robustness and uncertainty measures on challenging image classification benchmarks, closing the gap between previous methods and the best possible performance in some cases by more than half.

PDF Abstract ICLR 2020 PDF ICLR 2020 Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Domain Generalization ImageNet-C AugMix (ResNet-50) mean Corruption Error (mCE) 65.3 # 27
Domain Generalization ImageNet-R AugMix (ResNet-50) Top-1 Error Rate 58.9 # 25
Out-of-Distribution Generalization ImageNet-W AugMix (ResNet-50) IN-W Gap -16.8 # 1
Carton Gap +36 # 1
Out-of-Distribution Generalization UrbanCars AugMix BG Gap -10.3 # 1
CoObj Gap -12.1 # 1
BG+CoObj Gap -70.2 # 1