ITU

Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation

Research output: Conference Article in Proceeding or Book/Report chapterArticle in proceedingsResearchpeer-review

Standard

Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation. / Eskildsen, Anton Mølbjerg; Hansen, Dan Witzner.

ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications. Association for Computing Machinery, 2020. p. 1-9 5.

Research output: Conference Article in Proceeding or Book/Report chapterArticle in proceedingsResearchpeer-review

Harvard

Eskildsen, AM & Hansen, DW 2020, Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation. in ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications., 5, Association for Computing Machinery, pp. 1-9. https://doi.org/10.1145/3379155.3391327

APA

Eskildsen, A. M., & Hansen, D. W. (2020). Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation. In ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications (pp. 1-9). [5] Association for Computing Machinery. https://doi.org/10.1145/3379155.3391327

Vancouver

Eskildsen AM, Hansen DW. Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation. In ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications. Association for Computing Machinery. 2020. p. 1-9. 5 https://doi.org/10.1145/3379155.3391327

Author

Eskildsen, Anton Mølbjerg ; Hansen, Dan Witzner. / Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation. ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications. Association for Computing Machinery, 2020. pp. 1-9

Bibtex

@inproceedings{b9be64819342419b8f438d8400bd2811,
title = "Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation",
abstract = "We propose to use unlabelled eye image data for domain adaptation of an iris segmentation network. Adaptation allows the model to be less reliant on its initial generality. This is beneficial due to the large variance exhibited by eye image data which makes training of robust models difficult. The method uses a label prior in conjunction with network predictions to produce pseudo-labels. These are used in place of ground-truth data to adapt a base model. A fully connected neural network performs the pixel-wise iris segmentation. The base model is trained on synthetic data and adapted to several existing datasets with real-world eye images. The adapted models improve the average pupil centre detection rates by 24% at a distance of 25 pixels. We argue that the proposed method, and domain adaptation in general, is an interesting direction for increasing robustness of eye feature detectors. ",
author = "Eskildsen, {Anton M{\o}lbjerg} and Hansen, {Dan Witzner}",
year = "2020",
doi = "https://doi.org/10.1145/3379155.3391327",
language = "Dansk",
pages = "1--9",
booktitle = "ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications",
publisher = "Association for Computing Machinery",
address = "USA",

}

RIS

TY - GEN

T1 - Label Likelihood Maximisation: Adapting iris segmentation models using domain adaptation

AU - Eskildsen, Anton Mølbjerg

AU - Hansen, Dan Witzner

PY - 2020

Y1 - 2020

N2 - We propose to use unlabelled eye image data for domain adaptation of an iris segmentation network. Adaptation allows the model to be less reliant on its initial generality. This is beneficial due to the large variance exhibited by eye image data which makes training of robust models difficult. The method uses a label prior in conjunction with network predictions to produce pseudo-labels. These are used in place of ground-truth data to adapt a base model. A fully connected neural network performs the pixel-wise iris segmentation. The base model is trained on synthetic data and adapted to several existing datasets with real-world eye images. The adapted models improve the average pupil centre detection rates by 24% at a distance of 25 pixels. We argue that the proposed method, and domain adaptation in general, is an interesting direction for increasing robustness of eye feature detectors.

AB - We propose to use unlabelled eye image data for domain adaptation of an iris segmentation network. Adaptation allows the model to be less reliant on its initial generality. This is beneficial due to the large variance exhibited by eye image data which makes training of robust models difficult. The method uses a label prior in conjunction with network predictions to produce pseudo-labels. These are used in place of ground-truth data to adapt a base model. A fully connected neural network performs the pixel-wise iris segmentation. The base model is trained on synthetic data and adapted to several existing datasets with real-world eye images. The adapted models improve the average pupil centre detection rates by 24% at a distance of 25 pixels. We argue that the proposed method, and domain adaptation in general, is an interesting direction for increasing robustness of eye feature detectors.

U2 - https://doi.org/10.1145/3379155.3391327

DO - https://doi.org/10.1145/3379155.3391327

M3 - Konferencebidrag i proceedings

SP - 1

EP - 9

BT - ETRA '20 Full Papers: ACM Symposium on Eye Tracking Research and Applications

PB - Association for Computing Machinery

ER -

ID: 85227370