AutoAugment: Learning Augmentation Policies from Data

24 May 2018  ·  Ekin D. Cubuk, Barret Zoph, Dandelion Mane, Vijay Vasudevan, Quoc V. Le ·

Data augmentation is an effective technique for improving the accuracy of modern image classifiers. However, current data augmentation implementations are manually designed. In this paper, we describe a simple procedure called AutoAugment to automatically search for improved data augmentation policies. In our implementation, we have designed a search space where a policy consists of many sub-policies, one of which is randomly chosen for each image in each mini-batch. A sub-policy consists of two operations, each operation being an image processing function such as translation, rotation, or shearing, and the probabilities and magnitudes with which the functions are applied. We use a search algorithm to find the best policy such that the neural network yields the highest validation accuracy on a target dataset. Our method achieves state-of-the-art accuracy on CIFAR-10, CIFAR-100, SVHN, and ImageNet (without additional data). On ImageNet, we attain a Top-1 accuracy of 83.5% which is 0.4% better than the previous record of 83.1%. On CIFAR-10, we achieve an error rate of 1.5%, which is 0.6% better than the previous state-of-the-art. Augmentation policies we find are transferable between datasets. The policy learned on ImageNet transfers well to achieve significant improvements on other datasets, such as Oxford Flowers, Caltech-101, Oxford-IIT Pets, FGVC Aircraft, and Stanford Cars.

PDF Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Fine-Grained Image Classification Caltech-101 AutoAugment Top-1 Error Rate 13.07% # 11
Image Classification CIFAR-100 PyramidNet+ShakeDrop Percentage correct 89.3 # 33
Fine-Grained Image Classification FGVC Aircraft AutoAugment Top-1 Error Rate 7.33 # 2
Accuracy 92.67% # 37
Data Augmentation ImageNet ResNet-200 (AA) Accuracy (%) 80.0 # 6
Data Augmentation ImageNet ResNet-50 (AA) Accuracy (%) 77.6 # 12
Fine-Grained Image Classification Oxford 102 Flowers AutoAugment Top-1 Error Rate 4.64% # 4
Accuracy 95.36% # 22
Fine-Grained Image Classification Oxford-IIIT Pet Dataset AutoAugment Top-1 Error Rate 11.02% # 5
Accuracy 88.98% # 15
Fine-Grained Image Classification Stanford Cars AutoAugment Accuracy 94.8% # 25
Domain Generalization VizWiz-Classification EfficientNet-B1 (autoaug) Accuracy - All Images 39.7 # 38
Accuracy - Corrupted Images 32.8 # 39
Accuracy - Clean Images 44.4 # 38
Domain Generalization VizWiz-Classification EfficientNet-B2 (autoaug) Accuracy - All Images 41.6 # 28
Accuracy - Corrupted Images 34.3 # 32
Accuracy - Clean Images 45.8 # 28
Domain Generalization VizWiz-Classification EfficientNet-B0 (autoaug) Accuracy - All Images 34.9 # 72
Accuracy - Corrupted Images 27.3 # 78
Accuracy - Clean Images 40.1 # 64
Domain Generalization VizWiz-Classification EfficientNet-B3 (autoaug) Accuracy - All Images 42.6 # 22
Accuracy - Corrupted Images 34.9 # 29
Accuracy - Clean Images 47.5 # 20

Methods