Style transfer for generation of realistically textured subsurface models

Oleg Ovcharenko; Vladimir Kazei; Daniel Peter; Tariq Ali Alkhalifah

doi:10.1190/segam2019-3216349.1

Style transfer for generation of realistically textured subsurface models

Oleg Ovcharenko, Vladimir Kazei, Daniel Peter, Tariq Ali Alkhalifah

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

7 Scopus citations

Abstract

Training datasets consisting of numerous pairs of subsurface models and target variables are essential for building machine learning solutions for geophysical applications. We apply an iterative style transfer approach from image processing to produce realistically textured subsurface models based on synthetic prior models. The key idea of style transfer is that content and texture representations within a convolutional neural network are, to some extent, separable. Thus, a style from one image can be transferred to match the content from another image. We demonstrate examples where realistically random models are stylized to mimic texture patterns from Marmousi II and a section from the BP 2004 benchmark velocity models.

Original language	English (US)
Title of host publication	SEG Technical Program Expanded Abstracts 2019
Publisher	Society of Exploration Geophysicists
Pages	2393-2397
Number of pages	5
DOIs	https://doi.org/10.1190/segam2019-3216349.1
State	Published - Aug 10 2019

Bibliographical note

KAUST Repository Item: Exported on 2021-02-25
Acknowledgements: We thank Kevin Zakka for his implementation of the Gatys et al. (2015) algorithm (https://github.com/kevinzakka/style-transfer). The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST), Thuwal, 23955-6900, Saudi Arabia.

Access to Document

10.1190/segam2019-3216349.1

Cite this

@inproceedings{c9d062f7afe641cc83df066b7061d81c,

title = "Style transfer for generation of realistically textured subsurface models",

abstract = "Training datasets consisting of numerous pairs of subsurface models and target variables are essential for building machine learning solutions for geophysical applications. We apply an iterative style transfer approach from image processing to produce realistically textured subsurface models based on synthetic prior models. The key idea of style transfer is that content and texture representations within a convolutional neural network are, to some extent, separable. Thus, a style from one image can be transferred to match the content from another image. We demonstrate examples where realistically random models are stylized to mimic texture patterns from Marmousi II and a section from the BP 2004 benchmark velocity models.",

author = "Oleg Ovcharenko and Vladimir Kazei and Daniel Peter and Alkhalifah, {Tariq Ali}",

note = "KAUST Repository Item: Exported on 2021-02-25 Acknowledgements: We thank Kevin Zakka for his implementation of the Gatys et al. (2015) algorithm (https://github.com/kevinzakka/style-transfer). The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST), Thuwal, 23955-6900, Saudi Arabia.",

year = "2019",

month = aug,

day = "10",

doi = "10.1190/segam2019-3216349.1",

language = "English (US)",

pages = "2393--2397",

booktitle = "SEG Technical Program Expanded Abstracts 2019",

publisher = "Society of Exploration Geophysicists",

}

TY - GEN

T1 - Style transfer for generation of realistically textured subsurface models

AU - Ovcharenko, Oleg

AU - Kazei, Vladimir

AU - Peter, Daniel

AU - Alkhalifah, Tariq Ali

N1 - KAUST Repository Item: Exported on 2021-02-25 Acknowledgements: We thank Kevin Zakka for his implementation of the Gatys et al. (2015) algorithm (https://github.com/kevinzakka/style-transfer). The research reported in this publication was supported by funding from King Abdullah University of Science and Technology (KAUST), Thuwal, 23955-6900, Saudi Arabia.

PY - 2019/8/10

Y1 - 2019/8/10

N2 - Training datasets consisting of numerous pairs of subsurface models and target variables are essential for building machine learning solutions for geophysical applications. We apply an iterative style transfer approach from image processing to produce realistically textured subsurface models based on synthetic prior models. The key idea of style transfer is that content and texture representations within a convolutional neural network are, to some extent, separable. Thus, a style from one image can be transferred to match the content from another image. We demonstrate examples where realistically random models are stylized to mimic texture patterns from Marmousi II and a section from the BP 2004 benchmark velocity models.

AB - Training datasets consisting of numerous pairs of subsurface models and target variables are essential for building machine learning solutions for geophysical applications. We apply an iterative style transfer approach from image processing to produce realistically textured subsurface models based on synthetic prior models. The key idea of style transfer is that content and texture representations within a convolutional neural network are, to some extent, separable. Thus, a style from one image can be transferred to match the content from another image. We demonstrate examples where realistically random models are stylized to mimic texture patterns from Marmousi II and a section from the BP 2004 benchmark velocity models.

UR - http://hdl.handle.net/10754/667620

UR - https://library.seg.org/doi/10.1190/segam2019-3216349.1

UR - http://www.scopus.com/inward/record.url?scp=85079500110&partnerID=8YFLogxK

U2 - 10.1190/segam2019-3216349.1

DO - 10.1190/segam2019-3216349.1

M3 - Conference contribution

SP - 2393

EP - 2397

BT - SEG Technical Program Expanded Abstracts 2019

PB - Society of Exploration Geophysicists

ER -

Style transfer for generation of realistically textured subsurface models

Abstract

Bibliographical note

Access to Document

Other files and links

Fingerprint

Cite this