Meta Approach to Data Augmentation Optimization
This work addresses the need for efficient and scalable data augmentation optimization in image classification, though it is incremental as it builds on prior methods by avoiding proxy tasks.
The paper tackles the problem of optimizing data augmentation policies for image recognition tasks by simultaneously training models and policies via gradient descent, achieving improved performance on ImageNet and fine-grained recognition without dataset-specific tuning.
Data augmentation policies drastically improve the performance of image recognition tasks, especially when the policies are optimized for the target data and tasks. In this paper, we propose to optimize image recognition models and data augmentation policies simultaneously to improve the performance using gradient descent. Unlike prior methods, our approach avoids using proxy tasks or reducing search space, and can directly improve the validation performance. Our method achieves efficient and scalable training by approximating the gradient of policies by implicit gradient with Neumann series approximation. We demonstrate that our approach can improve the performance of various image classification tasks, including ImageNet classification and fine-grained recognition, without using dataset-specific hyperparameter tuning.