RandAugment: Practical automated data augmentation with a reduced search space

Recent work has shown that <PRE_TAG>data augmentation</POST_TAG> has the potential to significantly improve the <PRE_TAG>generalization</POST_TAG> of <PRE_TAG>deep learning models</POST_TAG>. Recently, <PRE_TAG>automated augmentation strategies</POST_TAG> have led to state-of-the-art results in image classification and <PRE_TAG>object detection</POST_TAG>. While these strategies were optimized for improving <PRE_TAG>validation accuracy</POST_TAG>, they also led to state-of-the-art results in <PRE_TAG>semi-supervised learning</POST_TAG> and improved <PRE_TAG>robustness to common corruptions</POST_TAG> of images. An obstacle to a large-scale adoption of these methods is a separate search phase which increases the training complexity and may substantially increase the computational cost. Additionally, due to the separate search phase, these approaches are unable to adjust the <PRE_TAG>regularization strength</POST_TAG> based on model or dataset size. Automated augmentation policies are often found by training small models on small datasets and subsequently applied to train larger models. In this work, we remove both of these obstacles. <PRE_TAG>RandAugment</POST_TAG> has a significantly reduced search space which allows it to be trained on the <PRE_TAG>target task</POST_TAG> with no need for a separate <PRE_TAG>proxy task</POST_TAG>. Furthermore, due to the parameterization, the <PRE_TAG>regularization strength</POST_TAG> may be tailored to different model and dataset sizes. <PRE_TAG>RandAugment</POST_TAG> can be used uniformly across different tasks and datasets and works out of the box, matching or surpassing all previous automated augmentation approaches on CIFAR-10/100, SVHN, and ImageNet. On the ImageNet dataset we achieve 85.0% accuracy, a 0.6% increase over the previous state-of-the-art and 1.0% increase over baseline augmentation. On <PRE_TAG>object detection</POST_TAG>, <PRE_TAG>RandAugment</POST_TAG> leads to 1.0-1.3% improvement over baseline augmentation, and is within 0.3% mAP of AutoAugment on COCO. Finally, due to its interpretable hyperparameter, <PRE_TAG>RandAugment</POST_TAG> may be used to investigate the role of <PRE_TAG>data augmentation</POST_TAG> with varying model and dataset size. Code is available online.