STaDA: Style Transfer as Data Augmentation

Xu, Zheng; Chalasani, Tejo; Ghosal, Koustav; Lutz, Sebastian; Smolić, Aljoša

doi:10.5220/0007353401070114

Cited by 27 publications

(19 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Many new advances in one style transfer field can inspire another style transfer field. For example, image style transfer has been used as a way for data augmentation (Zheng et al 2019;Jackson et al 2019) and adversarial attack , but TST has not yet been applied for such usage.…”

Section: Neural Style Transfermentioning

confidence: 99%

Deep Learning for Text Style Transfer: A Survey

Jin

et al. 2022

Computational Linguistics

View full text Add to dashboard Cite

Text style transfer is an important task in natural language generation, which aims to control certain attributes in the generated text, such as politeness, emotion, humor, and many others. It has a long history in the field of natural language processing, and recently has re-gained significant attention thanks to the promising performance brought by deep neural models. In this paper, we present a systematic survey of the research on neural text style transfer, spanning over 100 representative articles since the first neural text style transfer work in 2017. We discuss the task formulation, existing datasets and subtasks, evaluation, as well as the rich methodologies in the presence of parallel and non-parallel data. We also provide discussions on a variety of important topics regarding the future development of this task.

show abstract

Section: Neural Style Transfermentioning

confidence: 99%

Deep Learning for Text Style Transfer: A Survey

Jin

et al. 2022

Computational Linguistics

View full text Add to dashboard Cite

show abstract

“…Surprisingly, NST was not a subject of researches related to direct data augmentation until the end of 2019. The first experiment was to measure, if NST can improve a classifier's performance by creating stylised images, therefore the training dataset will have more variance, see paper [20] for further details. In 2020, NST was also used in medical fields for dermatological data augmentation, with the same motivations there as mentioned in the introduction, see paper [13].…”

Section: Related Workmentioning

confidence: 99%

Retinal Blood Vessel Segmentation on Style-augmented Images

Tóth

Kiss

2021

Studia UBB Informatica

View full text Add to dashboard Cite

The average human lifespan increased dramatically in the second half of 20th century. It was mainly due to technological improvements, which were driven by the continuous war preparations, and while humans have got another 20 years to live, unfortunately there are some sad side effects added to the elderly life. Various diseases can attack the eye, our major organ responsible for receiving information, therefore many researches were devoted to examine these diseases, their early signs, and how could they be stopped. From the start of 21th century, methods aided by computer were more and more involved in these processes, up to the current trend of using Convolutional Neural Networks (CNNs). While supervised methods, CNNs do achieve accuracy which can be compared to a skilled ophtalmologist, they require a tremendous amount of labeled data which is sparse in medical fields because the amount of time and resources needed to create them. One natural solution is to augment the data present, that is, copying the distribution while adding a small variety, like coloring an image differently. That is, what our paper aims to explore, whether a texturing algorithm, the Neural Style Transfery can be used to make a data set richer, and therefore helping a classifier CNN to achieve better results.

show abstract

“…Then it is used to relieve the domain shift issue soon. Some researchers used the generated images as data augmentation [25][26][27]. Moreover, image-to-image translation is used to transfer images to the target domain on which the downstream task model is trained [28,29].…”

Section: Unpaired Image-to-image Translationmentioning

confidence: 99%

Generative domain adaptation for chest X‐ray image analysis

Yin

Li²,

Fu³

et al. 2021

IET Image Processing

View full text Add to dashboard Cite

Chest X-ray images taken under different conditions follow different distributions, preventing the models trained on a domain from generalising well on the other domain. In this paper, a generative domain adaptation (GDA) method is proposed to address this issue and facilitate the learning process for downstream analysis. GDA adapts different domains to a virtual common one where images are aligned at the appearance level. To this end, a domain shared generator is used to transform the input images and two competitive discriminators are used to adversarially supervise the transforming process. The domain discriminator drives the generator to narrow the domain gap while the fidelity discriminator forces the generator to keep the inherent information. Moreover, a specific classification or detection network is attached to the generator to supervise it in a taskoriented manner. Experiment results on a large-scale dataset containing 46k chest X-ray images demonstrate that GDA outperforms representative domain adaptation methods by a large margin for both disease classification and lesion detection as well as provides useful transformed images to assist experts for diagnosis.How to cite this article: Yin, B., et al.: Generative domain adaptation for chest X-ray image analysis. IET Image Process. 15, 3118-3129 (2021).

show abstract

STaDA: Style Transfer as Data Augmentation

Cited by 27 publications

References 26 publications

Deep Learning for Text Style Transfer: A Survey

Deep Learning for Text Style Transfer: A Survey

Retinal Blood Vessel Segmentation on Style-augmented Images

Generative domain adaptation for chest X‐ray image analysis

Contact Info

Product

Resources

About