Challenges of Adversarial Image Augmentations

Arno Blaas, Xavier Suau, Jason Ramapuram, Nicholas Apostoloff, Luca Zappella
Proceedings on "I (Still) Can't Believe It's Not Better!" at NeurIPS 2021 Workshops, PMLR 163:9-14, 2022.

Abstract

Image augmentations applied during training are crucial for the generalization performance of image classifiers. Therefore, a large body of research has focused on finding the optimal augmentation policy for a given task. Yet, RandAugment [2], a simple random augmentation policy, has recently been shown to outperform existing sophisticated policies. Only Adversarial AutoAugment (AdvAA) [11], an approach based on the idea of adversarial training, has shown to be better than RandAugment. In this paper, we show that random augmentations are still competitive compared to an optimal adversarial approach, as well as to simple curricula, and conjecture that the success of AdvAA is due to the stochasticity of the policy controller network, which introduces a mild form of curriculum.

Cite this Paper

BibTeX
@InProceedings{pmlr-v163-blaas22a, title = {Challenges of Adversarial Image Augmentations}, author = {Blaas, Arno and Suau, Xavier and Ramapuram, Jason and Apostoloff, Nicholas and Zappella, Luca}, booktitle = {Proceedings on "I (Still) Can't Believe It's Not Better!" at NeurIPS 2021 Workshops}, pages = {9--14}, year = {2022}, editor = {Pradier, Melanie F. and Schein, Aaron and Hyland, Stephanie and Ruiz, Francisco J. R. and Forde, Jessica Z.}, volume = {163}, series = {Proceedings of Machine Learning Research}, month = {13 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v163/blaas22a/blaas22a.pdf}, url = {https://proceedings.mlr.press/v163/blaas22a.html}, abstract = {Image augmentations applied during training are crucial for the generalization performance of image classifiers. Therefore, a large body of research has focused on finding the optimal augmentation policy for a given task. Yet, RandAugment [2], a simple random augmentation policy, has recently been shown to outperform existing sophisticated policies. Only Adversarial AutoAugment (AdvAA) [11], an approach based on the idea of adversarial training, has shown to be better than RandAugment. In this paper, we show that random augmentations are still competitive compared to an optimal adversarial approach, as well as to simple curricula, and conjecture that the success of AdvAA is due to the stochasticity of the policy controller network, which introduces a mild form of curriculum.} }
Endnote
%0 Conference Paper %T Challenges of Adversarial Image Augmentations %A Arno Blaas %A Xavier Suau %A Jason Ramapuram %A Nicholas Apostoloff %A Luca Zappella %B Proceedings on "I (Still) Can't Believe It's Not Better!" at NeurIPS 2021 Workshops %C Proceedings of Machine Learning Research %D 2022 %E Melanie F. Pradier %E Aaron Schein %E Stephanie Hyland %E Francisco J. R. Ruiz %E Jessica Z. Forde %F pmlr-v163-blaas22a %I PMLR %P 9--14 %U https://proceedings.mlr.press/v163/blaas22a.html %V 163 %X Image augmentations applied during training are crucial for the generalization performance of image classifiers. Therefore, a large body of research has focused on finding the optimal augmentation policy for a given task. Yet, RandAugment [2], a simple random augmentation policy, has recently been shown to outperform existing sophisticated policies. Only Adversarial AutoAugment (AdvAA) [11], an approach based on the idea of adversarial training, has shown to be better than RandAugment. In this paper, we show that random augmentations are still competitive compared to an optimal adversarial approach, as well as to simple curricula, and conjecture that the success of AdvAA is due to the stochasticity of the policy controller network, which introduces a mild form of curriculum.
APA
Blaas, A., Suau, X., Ramapuram, J., Apostoloff, N. & Zappella, L.. (2022). Challenges of Adversarial Image Augmentations. Proceedings on "I (Still) Can't Believe It's Not Better!" at NeurIPS 2021 Workshops, in Proceedings of Machine Learning Research 163:9-14 Available from https://proceedings.mlr.press/v163/blaas22a.html.

Related Material