Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

You Only Need Half: Boosting Data Augmentation by Using Partial Content

Juntao Hu, Yuan Wu

TL;DR

You Only Need hAlf (YONA) targets robustness and generalization gaps in computer vision by a simple data augmentation: split an image into two halves, augment one half, mask the other, and reassemble. This patch-level, parameter-free approach reduces redundant information and encourages learning from incomplete views, while maintaining negligible computational cost. Empirical results show broad gains in CIFAR-10/100 across multiple architectures and augmentations, strong improvements in adversarial robustness under PGD, and strong ablation evidence that a 50% masking rate is optimal; however, benefits are task-dependent, with limited gains for object detection. The work suggests that partial-content training can complement existing augmentation policies and prompts a reevaluation of how image content is leveraged during augmentation for robust, generalizable models.

Abstract

We propose a novel data augmentation method termed You Only Need hAlf (YONA), which simplifies the augmentation process. YONA bisects an image, substitutes one half with noise, and applies data augmentation techniques to the remaining half. This method reduces the redundant information in the original image, encourages neural networks to recognize objects from incomplete views, and significantly enhances neural networks' robustness. YONA is distinguished by its properties of parameter-free, straightforward application, enhancing various existing data augmentation strategies, and thereby bolstering neural networks' robustness without additional computational cost. To demonstrate YONA's efficacy, extensive experiments were carried out. These experiments confirm YONA's compatibility with diverse data augmentation methods and neural network architectures, yielding substantial improvements in CIFAR classification tasks, sometimes outperforming conventional image-level data augmentation methods. Furthermore, YONA markedly increases the resilience of neural networks to adversarial attacks. Additional experiments exploring YONA's variants conclusively show that masking half of an image optimizes performance. The code is available at https://github.com/HansMoe/YONA.

You Only Need Half: Boosting Data Augmentation by Using Partial Content

TL;DR

You Only Need hAlf (YONA) targets robustness and generalization gaps in computer vision by a simple data augmentation: split an image into two halves, augment one half, mask the other, and reassemble. This patch-level, parameter-free approach reduces redundant information and encourages learning from incomplete views, while maintaining negligible computational cost. Empirical results show broad gains in CIFAR-10/100 across multiple architectures and augmentations, strong improvements in adversarial robustness under PGD, and strong ablation evidence that a 50% masking rate is optimal; however, benefits are task-dependent, with limited gains for object detection. The work suggests that partial-content training can complement existing augmentation policies and prompts a reevaluation of how image content is leveraged during augmentation for robust, generalizable models.

Abstract

We propose a novel data augmentation method termed You Only Need hAlf (YONA), which simplifies the augmentation process. YONA bisects an image, substitutes one half with noise, and applies data augmentation techniques to the remaining half. This method reduces the redundant information in the original image, encourages neural networks to recognize objects from incomplete views, and significantly enhances neural networks' robustness. YONA is distinguished by its properties of parameter-free, straightforward application, enhancing various existing data augmentation strategies, and thereby bolstering neural networks' robustness without additional computational cost. To demonstrate YONA's efficacy, extensive experiments were carried out. These experiments confirm YONA's compatibility with diverse data augmentation methods and neural network architectures, yielding substantial improvements in CIFAR classification tasks, sometimes outperforming conventional image-level data augmentation methods. Furthermore, YONA markedly increases the resilience of neural networks to adversarial attacks. Additional experiments exploring YONA's variants conclusively show that masking half of an image optimizes performance. The code is available at https://github.com/HansMoe/YONA.
Paper Structure (18 sections, 7 equations, 2 figures, 17 tables)

This paper contains 18 sections, 7 equations, 2 figures, 17 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Data Augmentation
  4. Patches in Vision
  5. Method
  6. Experiments
  7. Image Classification
  8. Object Detection
  9. Adversarial Robustness
  10. Ablation Study
  11. Conclusion
  12. Augmentations
  13. Object Detection
  14. Implementation Details
  15. Adversarial Robustness
  16. ...and 3 more sections

Figures (2)

  • Figure 1: The illustrations of the studied data augmentation approaches.
  • Figure 2: The illustrations of YONA operations. YONA cuts one image into two equal pieces, either in the height or the width dimension. A specific data augmentation method is performed on one piece, and the pixels within the other piece are replaced with noise. Transformed pieces are then concatenated together to form one single augmented image. The upper row shows the results of YONA applied with Cutout. The lower row presents the result of employing YONA to Vertical Flip.