Puzzle Mix: Exploiting Saliency and Local Statistics for Optimal Mixup
This work addresses the need for more effective data augmentation to improve model robustness and generalization in computer vision, representing an incremental advance over existing mixup methods.
The paper tackles the problem of overfitting and adversarial vulnerability in deep neural networks by proposing Puzzle Mix, a mixup augmentation method that leverages saliency and local statistics to create optimal virtual examples, achieving state-of-the-art generalization and adversarial robustness on datasets like CIFAR-100, Tiny-ImageNet, and ImageNet.
While deep neural networks achieve great performance on fitting the training distribution, the learned networks are prone to overfitting and are susceptible to adversarial attacks. In this regard, a number of mixup based augmentation methods have been recently proposed. However, these approaches mainly focus on creating previously unseen virtual examples and can sometimes provide misleading supervisory signal to the network. To this end, we propose Puzzle Mix, a mixup method for explicitly utilizing the saliency information and the underlying statistics of the natural examples. This leads to an interesting optimization problem alternating between the multi-label objective for optimal mixing mask and saliency discounted optimal transport objective. Our experiments show Puzzle Mix achieves the state of the art generalization and the adversarial robustness results compared to other mixup methods on CIFAR-100, Tiny-ImageNet, and ImageNet datasets. The source code is available at https://github.com/snu-mllab/PuzzleMix.