Improving Deep Learning with Generic Data Augmentation

By Luke Taylor, Geoff S. Nitschke. 2018

411 citations recorded by the source corpus.

Deep artificial neural networks require a large corpus of training data in order to effectively learn, where collection of such training data is often expensive and laborious.Data augmentationovercomes this issue by artificially inflating the training set with label preserving transformations. Recently there has been extensive use of generic data augmentation to improveConvolutional Neural Network(CNN) task performance. This study benchmarks various popular data augmentation schemes to allow researchers to make informed decisions as to which training methods are most appropriate for their data sets. Various geometric and photometric schemes are evaluated on a coarse-grained data set using a relatively simple CNN. Experimental results, run using 4-fold cross-validation and reported in terms of Top-1 and Top-5 accuracy, indicate that croppingin geometric augmentationsignificantly increases CNN task performance.

Subfields: Artificial Intelligence, 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 Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Albumentations: fast and flexible image augmentations (2018, background): “Although there have been attempts to evaluate individual contributions of image transform operations to overall accuracy [42], different augmentation strategies still lead to performance variability even on the well studied performance benchmarks.”
A Survey on Data Augmentation for Text Classification (2021, positive): “Data augmentation methods in computer vision are, among other, geometric transformations [7], [24], neural style transfers [31]–[33], interpolation of images [34], random partial deletions [35], and generative adversarial network (GAN) data generation [21].”
An Empirical Survey of Data Augmentation for Limited Data Learning in NLP (2021, methodology): “Such techniques have been widely explored in the computer vision ﬁeld, with methods like geometric/color space transformations (Simard et al., 2003; Krizhevsky et al., 2012; Taylor and Nitschke, 2018), mixup (Zhang et al., 2018), and random erasing (Zhong et al., 2020; DeVries and Taylor, 2017).”
The Role of AI in Drug Discovery: Challenges, Opportunities, and Strategies (2022, background): “This can increase the quantity and diversity of data available for training ML algorithms, improving the accuracy and reliability of the results 47 .”
Data Augmentation in Classification and Segmentation: A Survey and New Strategies (2023, background): “It is hypothesised that using training images with augmented edges could improve CNN performance, since the learned kernels in CNN could detect objects’ shapes more easily [47].”
A systematic review on overfitting control in shallow and deep neural networks (2021, methodology): “In a more different approach, the weights of Taylor and Nitschke (2018) Using geometric and photometric techniques to transform the shapes and shifting their pixels Zhong et al. (2020) Exchanging a random region of the input with some noise Cubuk et al. (2019) Using the reinforcement learning to…”
A Comprehensive Survey of Image Augmentation Techniques for Deep Learning (2022, methodology): “To address this issue, image augmentation has been conﬁrmed to be an eﬀective and eﬃcient strategy [3, 4].”