RandAugment: Practical automated data augmentation with a reduced search\n space

2019/09/30 by Ekin D. Cubuk, Cubuk, Ekin D., Barret Zoph +5 · 114 citations
Computer Science · #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Multimodal Machine Learning Applications

paper · pdf · doi:10.48550/arxiv.1909.13719

openalex publication_date 2019/09/30 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Recent work has shown that data augmentation has the potential to\nsignificantly improve the generalization of deep learning models. Recently,\nautomated augmentation strategies have led to state-of-the-art results in image\nclassification and object detection. While these strategies were optimized for\nimproving validation accuracy, they also led to state-of-the-art results in\nsemi-supervised learning and improved robustness to common corruptions of\nimages. An obstacle to a large-scale adoption of these methods is a separate\nsearch phase which increases the training complexity and may substantially\nincrease the computational cost. Additionally, due to the separate search\nphase, these approaches are unable to adjust the regularization strength based\non model or dataset size. Automated augmentation policies are often found by\ntraining small models on small datasets and subsequently applied to train\nlarger models. In this work, we remove both of these obstacles. RandAugment has\na significantly reduced search space which allows it to be trained on the\ntarget task with no need for a separate proxy task. Furthermore, due to the\nparameterization, the regularization strength may be tailored to different\nmodel and dataset sizes. RandAugment can be used uniformly across different\ntasks and datasets and works out of the box, matching or surpassing all\nprevious automated augmentation approaches on CIFAR-10/100, SVHN, and ImageNet.\nOn the ImageNet dataset we achieve 85.0% accuracy, a 0.6% increase over the\nprevious state-of-the-art and 1.0% increase over baseline augmentation. On\nobject detection, RandAugment leads to 1.0-1.3% improvement over baseline\naugmentation, and is within 0.3% mAP of AutoAugment on COCO. Finally, due to\nits interpretable hyperparameter, RandAugment may be used to investigate the\nrole of data augmentation with varying model and dataset size. Code is\navailable online.\n

Cited by

• Multi-label Classification with Panoptic Context Aggregation Networks
• AGVBench: A Reliability-Oriented Benchmark of Data Augmentation for Vein Recognition
• IJCB-AFMFR 2026: Competition on Adapting Foundation Models for Face Recognition Using Synthetic Training Data
• Vision Transformers are Circulant Attention Learners
• UniMPR: A Unified Framework for Multimodal Place Recognition with Heterogeneous Sensor Configurations
• Next-Embedding Prediction Makes Strong Vision Learners
• Interaction-via-Actions: Cattle Interaction Detection with Joint Learning of Action-Interaction Latent Space
• Mimicking Human Visual Development for Learning Robust Image Representations
• Quality-Driven and Diversity-Aware Sample Expansion for Robust Marine Obstacle Segmentation
• GrowTAS: Progressive Expansion from Small to Large Subnets for Efficient ViT Architecture Search
• Contrastive Time Series Forecasting with Anomalies
• Weakly Supervised Tuberculosis Localization in Chest X-rays through Knowledge Distillation
• Leveraging Text Guidance for Enhancing Demographic Fairness in Gender Classification
• Correction of Decoupled Weight Decay
• A Comparative Study on Synthetic Facial Data Generation Techniques for Face Recognition
• Label-Efficient Point Cloud Segmentation with Active Learning
• LeAD-M3D: Leveraging Asymmetric Distillation for Real-Time Monocular 3D Detection
• Parabolic Position Encoding: Vision-Centric, Principled, Extrapolatable, General
• Rethinking Decoupled Knowledge Distillation: A Predictive Distribution Perspective
• A Retrieval-Augmented Generation Approach to Extracting Algorithmic Logic from Neural Networks
• Multi-Scale Visual Prompting for Lightweight Small-Image Classification
• ViT³: Unlocking Test-Time Training in Vision
• InternVideo-Next: Towards General Video Foundation Models without Video-Text Supervision
• ShelfRectNet: Single View Shelf Image Rectification with Homography Estimation
• Advancing Image Classification with Discrete Diffusion Classification Modeling
• DiCaP: Distribution-Calibrated Pseudo-labeling for Semi-Supervised Multi-Label Learning
• On the Role of Hidden States of Modern Hopfield Network in Transformer
• Rethinking Vision Transformer Depth via Structural Reparameterization
• Unifying Convolution and Attention via Convolutional Nearest Neighbors
• Semi-Supervised Multi-Task Learning for Interpretable Quality As- sessment of Fundus Images
• HiGFA: Hierarchical Guidance for Fine-grained Data Augmentation with Diffusion Models
• Learning Time in Static Classifiers
• D-GAP: Improving Out-of-Domain Robustness via Dataset-Agnostic and Gradient-Guided Augmentation in Amplitude and Pixel Spaces
• CATCHFed: Efficient Unlabeled Data Utilization for Semi-Supervised Federated Learning in Limited Labels Environments
• Intrinsic Dimensionality as a Model-Free Measure of Class Imbalance
• AdaptViG: Adaptive Vision GNN with Exponential Decay Gating
• Do Blind Spots Matter for Word-Referent Mapping? A Computational Study with Infant Egocentric Video
• FedeCouple: Fine-Grained Balancing of Global-Generalization and Local-Adaptability in Federated Learning
• Iterated Population Based Training with Task-Agnostic Restarts
• Linear Differential Vision Transformer: Learning Visual Contrasts via Pairwise Differentials
• Gradient-Weight Alignment as a Train-Time Proxy for Generalization in Classification Tasks
• Searching for Robust Augmentations to Improve Out-of-Domain Generalization in Dermoscopic Skin Cancer Classification
• UHKD: A Unified Framework for Heterogeneous Knowledge Distillation via Frequency-Domain Representations
• Humanoid-inspired Causal Representation Learning for Domain Generalization
• MergeMix: A Unified Augmentation Paradigm for Visual and Multi-Modal Understanding
• SemiETPicker: Fast and Label-Efficient Particle Picking for CryoET Tomography Using Semi-Supervised Learning
• What Does It Take to Build a Performant Selective Classifier?
• Exploring "Many in Few" and "Few in Many" Properties in Long-Tailed, Highly-Imbalanced IC Defect Classification
• Accelerating Vision Transformers with Adaptive Patch Sizes
• Beyond Real Faces: Synthetic Datasets Can Achieve Reliable Recognition Performance without Privacy Compromise
• ObjectTransforms for Uncertainty Quantification and Reduction in Vision-Based Perception for Autonomous Vehicles
• ReCon: Region-Controllable Data Augmentation with Rectification and Alignment for Object Detection
• Salient Concept-Aware Generative Data Augmentation
• Multi-Scale High-Resolution Logarithmic Grapher Module for Efficient Vision GNNs
• Assessing the Potential for Catastrophic Failure in Dynamic Post-Training Quantization
• Universal Adaptive Environment Discovery
• Class Prototypes based Contrastive Learning for Classifying Multi-Label and Fine-Grained Educational Videos
• Test-Time Adaptation by Causal Trimming
• microCLIP: Unsupervised CLIP Adaptation via Coarse-Fine Token Fusion for Fine-Grained Image Classification
• Task-Aware Resolution Optimization for Visual Large Language Models
• Beyond Real Data: Synthetic Data through the Lens of Regularization
• Enhancing Visual Prompting through Expanded Transformation Space and Overfitting Mitigation
• Evaluating Fundus-Specific Foundation Models for Diabetic Macular Edema Detection
• Revisiting Mixout: An Overlooked Path to Robust Finetuning
• High-Rate Mixout: Revisiting Mixout for Robust Domain Generalization
• Using predefined vector systems as latent space configuration for neural network supervised training on data with arbitrarily large number of classes
• Towards Adversarial Training under Hyperspectral Images
• On-the-Fly Data Augmentation via Gradient-Guided and Sample-Aware Influence Estimation
• OmniDFA: A Unified Framework for Open Set Synthesis Image Detection and Few-Shot Attribution
• LABELING COPILOT: A Deep Research Agent for Automated Data Curation in Computer Vision
• SpikeMatch: Semi-Supervised Learning with Temporal Dynamics of Spiking Neural Networks
• Unlocking Noise-Resistant Vision: Key Architectural Secrets for Robust Models
• Biologically Plausible Learning via Bidirectional Spike-Based Distillation
• What Makes Deep Learning Work for Traditional Chinese Medicine Tongue Diagnosis? A Comprehensive Ablation Study
• ViG-LRGC: Vision Graph Neural Networks with Learnable Reparameterized Graph Construction
• Intra-Cluster Mixup: An Effective Data Augmentation Technique for Complementary-Label Learning
• Multimodal Medical Image Classification via Synergistic Learning Pre-training
• IPF-RDA: An Information-Preserving Framework for Robust Data Augmentation
• ToFU: Transforming How Federated Learning Systems Forget User Data
• An Exploratory Study on Abstract Images and Visual Representations Learned from Them
• Self Identity Mapping
• The Quest for Universal Master Key Filters in DS-CNNs
• Adversarial robustness through Lipschitz-Guided Stochastic Depth in Neural Networks
• CoSwin: Convolution Enhanced Hierarchical Shifted Window Attention For Small-Scale Vision
• Rethinking the Backbone in Class Imbalanced Federated Source Free Domain Adaptation: The Utility of Vision Foundation Models
• Lookup multivariate Kolmogorov-Arnold Networks
• VCMamba: Bridging Convolutions with Multi-Directional Mamba for Efficient Visual Representation
• Data-Augmented Quantization-Aware Knowledge Distillation
• STA-Net: A Decoupled Shape and Texture Attention Network for Lightweight Plant Disease Classification
• Enhancing Zero-Shot Pedestrian Attribute Recognition with Synthetic Data Generation: A Comparative Study with Image-To-Image Diffusion Models
• A Data-Centric Approach to Pedestrian Attribute Recognition: Synthetic Augmentation via Prompt-driven Diffusion Models
• Expandable Residual Approximation for Knowledge Distillation
• A Two-Stage Strategy for Mitosis Detection Using Improved YOLO11x Proposals and ConvNeXt Classification
• Feed Two Birds with One Scone: Exploiting Function-Space Regularization for Both OOD Robustness and ID Fine-Tuning Performance
• Target-Oriented Single Domain Generalization
• E-ConvNeXt: A Lightweight and Efficient ConvNeXt Variant with Cross-Stage Partial Connections
• From Linearity to Non-Linearity: How Masked Autoencoders Capture Spatial Correlations
• Data Augmentation Improves Machine Unlearning
• Learning Spatial Decay for Vision Transformers
• UniConvNet: Expanding Effective Receptive Field while Maintaining Asymptotically Gaussian Distribution for ConvNets of Any Scale
• Exploring Cross-Stage Adversarial Transferability in Class-Incremental Continual Learning
• Multi-level Collaborative Distillation Meets Global Workspace Model: A Unified Framework for OCIL
• Toward Robust Semi-supervised Regression via Dual-stream Knowledge Distillation
• Sample-aware RandAugment: Search-free Automatic Data Augmentation for Effective Image Recognition
• Enhancing Small-Scale Dataset Expansion with Triplet-Connection-based Sample Re-Weighting
• TSMS-SAM2: Multi-scale Temporal Sampling Augmentation and Memory-Splitting Pruning for Promptable Video Object Segmentation and Tracking in Surgical Scenarios
• The Power of Many: Synergistic Unification of Diverse Augmentations for Efficient Adversarial Robustness
• ε-Softmax: Approximating One-Hot Vectors for Mitigating Label Noise
• Raw Data Matters: Enhancing Prompt Tuning by Internal Augmentation on Vision-Language Models
• Towards Robust Semantic Correspondence: A Benchmark and Insights
• Analysis of Hyperparameter Optimization Effects on Lightweight Deep Models for Real-Time Image Classification
• Multi-Modal Motion Retrieval by Learning a Fine-Grained Joint Embedding Space
• CaliMatch: Adaptive Calibration for Improving Safe Semi-supervised Learning
• A Semi-Supervised Federated Learning Framework with Hierarchical Clustering Aggregation for Heterogeneous Satellite Networks

Related

• Sample-aware RandAugment: Search-free Automatic Data Augmentation for Effective Image Recognition
• AutoAugment: Learning Augmentation Policies from Data
• UniformAugment: A Search-free Probabilistic Data Augmentation Approach
• Scale-aware Automatic Augmentation for Object Detection
• RandMSAugment: A Mixed-Sample Augmentation for Limited-Data Scenarios