The Effectiveness of Data Augmentation in Image Classification using Deep Learning

By Luis Perez, Jason Wang. 2017

3,170 citations recorded by the source corpus.

In this paper, we explore and compare multiple solutions to the problem of data augmentation in image classification. Previous work has demonstrated the effectiveness of data augmentation through simple techniques, such as cropping, rotating, and flipping input images. We artificially constrain our access to data to a small subset of the ImageNet dataset, and compare each data augmentation technique in turn. One of the more successful data augmentations strategies is the traditional transformations mentioned above. We also experiment with GANs to generate images of different styles. Finally, we propose a method to allow a neural net to learn augmentations that best improve the classifier, which we call neural augmentation. We discuss the successes and shortcomings of this method on various datasets.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

Randaugment: Practical automated data augmentation with a reduced search space (2019, methodology): “GANs have also been used to generate training data directly [37, 33, 56, 1, 44], however this approach does not seem to be as beneﬁcial as learning sequences of data augmentation operations that are pre-deﬁned [40].”
AutoAugment: Learning Augmentation Policies from Data (2018, methodology): “Generative adversarial networks have also been used for the purpose of generating additional data (e.g., [45–49]).”
Theoretical evidence for adversarial robustness through randomization: the case of the Exponential family (2019, methodology): “robust optimization [4] and data augmentation techniques [36].”
Real-Time Federated Evolutionary Neural Architecture Search (2020, neutral): “Note that we do not apply any data augmentation [57] in our simulation, since the server cannot do any operations on the client data in federated learning.”
Long-Tailed Class Incremental Learning (2022, background): “Data augmentation is another way to increase the tail distribution [30].”
GAN Memory with No Forgetting (2020, background): “…consequently also protecting privacy (the original data need not be saved); and ( iii ) a GAN often generates realistic samples not observed in training data, delivering a synthetic data augmentation that potentially beneﬁts better performance of downstream tasks [78, 8, 9, 21, 33, 26, 27].”
A survey on Image Data Augmentation for Deep Learning (2019, methodology): “Perez and Wang tested their algorithm on the MNIST and Tiny-imagenet-200 datasets on binary classification tasks such as cat versus dog.”
MixText: Linguistically-Informed Interpolation of Hidden Space for Semi-Supervised Text Classification (2020, background): “For instance, in computer vision, images are shifted, zoomed in/out, rotated, ﬂipped, distorted, or shaded with a hue (Perez and Wang, 2017) for training data augmentation.”
Data Augmentation by Pairing Samples for Images Classification (2018, methodology): “The basic idea of our technique is simple: We synthesize a new image from two images randomly picked from the training set as done in some existing techniques, such as Chawla et al. (2002) and Wang & Perez (2017).”
Neural Data Augmentation via Example Extrapolation (2021, methodology): “Data augmentation is a popular solution for biased or imbalanced data, either by duplicating examples or using heuristics to synthesize new examples (Perez and Wang, 2017).”
Learning Data Augmentation Strategies for Object Detection (2019, methodology): “For image classification models, data can be augmented either by learning a generator that can create data from scratch [33, 31, 54, 2, 43], or by learning a set of transformations as applied to already existing training set samples [5, 37].”
Label Refinery: Improving ImageNet Classification through Label Progression (2018, methodology): “A variety of data augmentation methods [1,2,3,4] have effectively expanded these datasets and improved the performance of learning systems.”
G-DAug: Generative Data Augmentation for Commonsense Reasoning (2020, background): “puter vision, where it takes the form of image transformations like translation and rotation (Perez and Wang, 2017).”
Distribution Augmentation for Low-Resource Expressive Text-To-Speech (2022, neutral): “Data augmentation techniques play an important role in the training of neural networks, particularly in classification tasks [1, 2, 3].”