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

Abstract

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.

Code available at https://github.com/google-research/augmix

References in corpus (7)

• Improved Regularization of Convolutional Neural Networks with Cutout
• On Calibration of Modern Neural Networks
• Random Erasing Data Augmentation
• Deep Anomaly Detection with Outlier Exposure
• Making Convolutional Networks Shift-Invariant Again
• Learning with Pseudo-Ensembles
• Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation

Cited by in corpus (26)

• MedViT: A Robust Vision Transformer for Generalized Medical Image Classification
• An Overview of Deep Semi-Supervised Learning
• Robust Pre-Training by Adversarial Contrastive Learning
• Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformation
• ResNet strikes back: An improved training procedure in timm
• Assaying Out-Of-Distribution Generalization in Transfer Learning
• TGDM: Target Guided Dynamic Mixup for Cross-Domain Few-Shot Learning
• If a Human Can See It, So Should Your System: Reliability Requirements for Machine Vision Components
• Class Gradient Projection For Continual Learning
• Augmentation Strategies for Learning with Noisy Labels
• DeepLens: Interactive Out-of-distribution Data Detection in NLP Models
• A Self-Supervised Feature Map Augmentation (FMA) Loss and Combined Augmentations Finetuning to Efficiently Improve the Robustness of CNNs
• ObjectAug: Object-level Data Augmentation for Semantic Image Segmentation
• A Decoupled Uncertainty Model for MRI Segmentation Quality Estimation
• A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution Robustness
• Learning Temporally Invariant and Localizable Features via Data Augmentation for Video Recognition
• The Connection between Out-of-Distribution Generalization and Privacy of ML Models
• Random Shadows and Highlights: A new data augmentation method for extreme lighting conditions
• Extending Label Smoothing Regularization with Self-Knowledge Distillation
• Noise Optimization for Artificial Neural Networks
• Estimating Predictive Uncertainty Under Program Data Distribution Shift
• SuperpixelGridCut, SuperpixelGridMean and SuperpixelGridMix Data Augmentation
• Online Black-Box Confidence Estimation of Deep Neural Networks
• Untapped Potential of Data Augmentation: A Domain Generalization Viewpoint
• Whole Slide Multiple Instance Learning for Predicting Axillary Lymph Node Metastasis
• Nuisance-Label Supervision: Robustness Improvement by Free Labels