A Study of Full Stage Data Augmentation Method
Keywords:
computer vision, deep learning, data augmentation, model ensemble, training data, testing data, geometric transformations, color space augmentations, kernel filters, picture mixing, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, meta-learning, model performance, big dataAbstract
Many computer vision-related tasks may now be accomplished using deep learning, which relies heavily on the use of huge data. A full-stage data augmentation approach for deep learning is proposed in this, which may also serve as an implicit model ensemble without incurring additional model training expenses. A data-space solution to the problem of limited data can be achieved by simultaneously augmenting training and testing data. Deep Learning models may be developed utilizing larger and more accurate training datasets that have been enhanced by data augmentation approaches. Geometric transformations, color space augmentations, kernel filters, picture mixing, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning are among the image augmentation methods explored in this review. Deep Learning models may be developed utilizing larger and more accurate training datasets thanks to Data Augmentation approaches. Improve model performance and increase datasets to take use of big data's possibilities through data augmentation.Downloads
Download data is not yet available.
Published
2019-01-01
Issue
Section
Articles
How to Cite
[1]
“A Study of Full Stage Data Augmentation Method”, JASRAE, vol. 16, no. 1, pp. 3214–3218, Jan. 2019, Accessed: Nov. 21, 2026. Available: https://www.ignited.in/index.php/jasrae/article/view/10042