Understanding the Class-Specific Effects of Data Augmentations

Abstract

Data augmentation (DA) is a major part of modern computer vision used to encode invariance and improve generalization. However, recent studies have shown that the effects of DA can be highly class dependent: augmentation strategies that improve average accuracy may significantly hurt the accuracies on a minority of individual classes, e.g. by as much as $20\%$ on ImageNet. In this work, we explain this phenomenon from the perspective of interactions among class-conditional distributions. We find that most affected classes are inherently ambiguous, co-occur, or involve fine-grained distinctions. By using the higher-quality multi-label ImageNet annotations, we show the negative effects of data augmentation on per-class accuracy are significantly less severe.

Cite

Text

Kirichenko et al. "Understanding the Class-Specific Effects of Data Augmentations." ICLR 2023 Workshops: Trustworthy_ML, 2023.

Markdown

[Kirichenko et al. "Understanding the Class-Specific Effects of Data Augmentations." ICLR 2023 Workshops: Trustworthy_ML, 2023.](https://mlanthology.org/iclrw/2023/kirichenko2023iclrw-understanding/)

BibTeX

@inproceedings{kirichenko2023iclrw-understanding,
  title     = {{Understanding the Class-Specific Effects of Data Augmentations}},
  author    = {Kirichenko, Polina and Balestriero, Randall and Ibrahim, Mark and Vedantam, Shanmukha Ramakrishna and Firooz, Hamed and Wilson, Andrew Gordon},
  booktitle = {ICLR 2023 Workshops: Trustworthy_ML},
  year      = {2023},
  url       = {https://mlanthology.org/iclrw/2023/kirichenko2023iclrw-understanding/}
}