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 (26)

• mixup: Beyond Empirical Risk Minimization
• Deep Residual Learning for Image Recognition
• Improved Regularization of Convolutional Neural Networks with Cutout
• Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour
• On Calibration of Modern Neural Networks
• Unsupervised Data Augmentation for Consistency Training
• A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks
• Random Erasing Data Augmentation
• AutoAugment: Learning Augmentation Policies from Data
• CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features
• Data Augmentation using Random Image Cropping and Patching for Deep CNNs
• Using Pre-Training Can Improve Model Robustness and Uncertainty
• Deep Anomaly Detection with Outlier Exposure
• Aggregated Residual Transformations for Deep Neural Networks
• Making Convolutional Networks Shift-Invariant Again
• Learning with Pseudo-Ensembles
• Why do deep convolutional networks generalize so poorly to small image transformations?
• Motivating the Rules of the Game for Adversarial Example Research
• Examining the Impact of Blur on Recognition by Convolutional Networks
• Adversarial Training Can Hurt Generalization
• Detecting and Correcting for Label Shift with Black Box Predictors
• Improving Robustness Without Sacrificing Accuracy with Patch Gaussian Augmentation
• A Fourier Perspective on Model Robustness in Computer Vision
• Evaluating and Understanding the Robustness of Adversarial Logit Pairing
• Testing Robustness Against Unforeseen Adversaries
• An Empirical Evaluation on Robustness and Uncertainty of Regularization Methods

Cited by in corpus (76)

• A Review of Uncertainty Quantification in Deep Learning: Techniques, Applications and Challenges
• Albumentations: fast and flexible image augmentations
• FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
• MedViT: A Robust Vision Transformer for Generalized Medical Image Classification
• A Survey on Data Augmentation for Text Classification
• A Simple Semi-Supervised Learning Framework for Object Detection
• Generalized Out-of-Distribution Detection: A Survey
• An Overview of Deep Semi-Supervised Learning
• Perceiver: General Perception with Iterative Attention
• FMix: Enhancing Mixed Sample Data Augmentation
• Evaluating Prediction-Time Batch Normalization for Robustness under Covariate Shift
• An overview of mixing augmentation methods and augmentation strategies
• Robust Pre-Training by Adversarial Contrastive Learning
• A Survey of Deep Learning for Scientific Discovery
• DrFuse: Learning Disentangled Representation for Clinical Multi-Modal Fusion with Missing Modality and Modal Inconsistency
• Greedy Policy Search: A Simple Baseline for Learnable Test-Time Augmentation
• Learning Invariance from Generated Variance for Unsupervised Person Re-identification
• Learning perturbation sets for robust machine learning
• Test-Time Adaptation to Distribution Shift by Confidence Maximization and Input Transformation
• PatchUp: A Feature-Space Block-Level Regularization Technique for Convolutional Neural Networks
• ResNet strikes back: An improved training procedure in timm
• Meta-FDMixup: Cross-Domain Few-Shot Learning Guided by Labeled Target Data
• Assaying Out-Of-Distribution Generalization in Transfer Learning
• TGDM: Target Guided Dynamic Mixup for Cross-Domain Few-Shot Learning
• GuidedMixup: An Efficient Mixup Strategy Guided by Saliency Maps
• If a Human Can See It, So Should Your System: Reliability Requirements for Machine Vision Components
• On the effectiveness of adversarial training against common corruptions
• Fourier-basis Functions to Bridge Augmentation Gap: Rethinking Frequency Augmentation in Image Classification
• Transfer Learning Gaussian Anomaly Detection by Fine-tuning Representations
• Diffusion Based Augmentation for Captioning and Retrieval in Cultural Heritage
• Class Gradient Projection For Continual Learning
• Robust Image Classification Using A Low-Pass Activation Function and DCT Augmentation
• Understanding Generalization in Adversarial Training via the Bias-Variance Decomposition
• Towards Robust Vision Transformer
• Augmentation Strategies for Learning with Noisy Labels
• Geometric Understanding of Discriminability and Transferability for Visual Domain Adaptation
• Combining Ensembles and Data Augmentation can Harm your Calibration
• It Takes Two to Tango: Mixup for Deep Metric Learning
• DeepLens: Interactive Out-of-distribution Data Detection in NLP Models
• A Study on the Impact of Data Augmentation for Training Convolutional Neural Networks in the Presence of Noisy Labels
• When Human Pose Estimation Meets Robustness: Adversarial Algorithms and Benchmarks
• Model-Based Domain Generalization
• ReSmooth: Detecting and Utilizing OOD Samples when Training with Data Augmentation
• ObjectAug: Object-level Data Augmentation for Semantic Image Segmentation
• A Self-Supervised Feature Map Augmentation (FMA) Loss and Combined Augmentations Finetuning to Efficiently Improve the Robustness of CNNs
• Does Data Augmentation Benefit from Split BatchNorms
• A Winning Hand: Compressing Deep Networks Can Improve Out-Of-Distribution Robustness
• A Decoupled Uncertainty Model for MRI Segmentation Quality Estimation
• On Calibration of Mixup Training for Deep Neural Networks
• A Pairwise DomMix Attentive Adversarial Network for Unsupervised Domain Adaptive Object Detection
• Self-Supervised Place Recognition by Refining Temporal and Featural Pseudo Labels from Panoramic Data
• Model-Based Robust Deep Learning: Generalizing to Natural, Out-of-Distribution Data
• Learning Temporally Invariant and Localizable Features via Data Augmentation for Video Recognition
• Continual Learning in the Presence of Repetition
• Assessment of the Reliablity of a Model's Decision by Generalizing Attribution to the Wavelet Domain
• Causality-inspired Single-source Domain Generalization for Medical Image Segmentation
• 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
• Estimating Predictive Uncertainty Under Program Data Distribution Shift
• Noise Optimization for Artificial Neural Networks
• Extending Label Smoothing Regularization with Self-Knowledge Distillation
• Improving Facial Emotion Recognition through Dataset Merging and Balanced Training Strategies
• Untapped Potential of Data Augmentation: A Domain Generalization Viewpoint
• ParamCrop: Parametric Cubic Cropping for Video Contrastive Learning
• SPADE: A Spectral Method for Black-Box Adversarial Robustness Evaluation
• Golden Grain: Building a Secure and Decentralized Model Marketplace for MLaaS
• DeepMix: Online Auto Data Augmentation for Robust Visual Object Tracking
• SuperpixelGridCut, SuperpixelGridMean and SuperpixelGridMix Data Augmentation
• Online Black-Box Confidence Estimation of Deep Neural Networks
• Whole Slide Multiple Instance Learning for Predicting Axillary Lymph Node Metastasis
• The Uncanny Similarity of Recurrence and Depth
• Gradient-based Data Augmentation for Semi-Supervised Learning
• Noisy Label Classification using Label Noise Selection with Test-Time Augmentation Cross-Entropy and NoiseMix Learning
• Noisy Feature Mixup
• What Can Style Transfer and Paintings Do For Model Robustness?
• Nuisance-Label Supervision: Robustness Improvement by Free Labels