PTGaze: Cross-domain gaze estimation via proxy tuning

Yafei Wang; Runze Yan; Yaxiong Lei; Xianping Fu

doi:10.1145/3715669.372682

PTGaze: Cross-domain gaze estimation via proxy tuning

Yafei Wang, Runze Yan, Yaxiong Lei, Xianping Fu

School of Computer Science

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

Abstract

To tackle the cross-domain model performance degradation challenge, a gaze estimation method based on proxy tuning is proposed, called PTGaze. In PTGaze, an base model is used to learn the gaze representation of the baseline model in the source domain, and an adapt model is used to learn the gaze representation of the baseline model in the target domain. The gaze difference between the base and adapt models is utilized to guide the final output, ensuring the method’s accuracy in the target domain. Experimental results show that the proposed method achieves higher cross-domain gaze estimation accuracy on five public datasets, using RT-Gene, Full-face, and Gaze360 as baseline models.

Original language	English
Title of host publication	Proceedings of the 2025 symposium on eye tracking research and applications (ETRA)
Editors	Yusuke Sugano, Mohamed Khamis, Aladine Chetouani, Ludwig Sidenmark, Alessandro Bruno
Place of Publication	New York
Publisher	ACM
Pages	1-2
Number of pages	2
ISBN (Print)	9798400714870
DOIs	https://doi.org/10.1145/3715669.372682
Publication status	Published - 25 May 2025

Access to Document

10.1145/3715669.372682

Cite this

@inproceedings{fb376b5026174e7982879570f1923db0,

title = "PTGaze: Cross-domain gaze estimation via proxy tuning",

abstract = "To tackle the cross-domain model performance degradation challenge, a gaze estimation method based on proxy tuning is proposed, called PTGaze. In PTGaze, an base model is used to learn the gaze representation of the baseline model in the source domain, and an adapt model is used to learn the gaze representation of the baseline model in the target domain. The gaze difference between the base and adapt models is utilized to guide the final output, ensuring the method{\textquoteright}s accuracy in the target domain. Experimental results show that the proposed method achieves higher cross-domain gaze estimation accuracy on five public datasets, using RT-Gene, Full-face, and Gaze360 as baseline models.",

author = "Yafei Wang and Runze Yan and Yaxiong Lei and Xianping Fu",

note = "Funding: This work was supported in part by the National Key Research and Development Program of China Grant 2022YFE0132600, by the National Natural Science Foundation of China Grants 62002043, by the Research Project of China Disabled Persons{\textquoteright} Federation - on Assistive Technology Grant 2024CDPFAT-28.",

year = "2025",

month = may,

day = "25",

doi = "10.1145/3715669.372682",

language = "English",

isbn = "9798400714870",

pages = "1--2",

editor = "Yusuke Sugano and Mohamed Khamis and Aladine Chetouani and Ludwig Sidenmark and Alessandro Bruno",

booktitle = "Proceedings of the 2025 symposium on eye tracking research and applications (ETRA)",

publisher = "ACM",

address = "United States",

}

PTGaze: Cross-domain gaze estimation via proxy tuning. / Wang, Yafei; Yan, Runze; Lei, Yaxiong et al.
Proceedings of the 2025 symposium on eye tracking research and applications (ETRA). ed. / Yusuke Sugano; Mohamed Khamis; Aladine Chetouani; Ludwig Sidenmark; Alessandro Bruno. New York: ACM, 2025. p. 1-2 60.

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

TY - GEN

T1 - PTGaze: Cross-domain gaze estimation via proxy tuning

AU - Wang, Yafei

AU - Yan, Runze

AU - Lei, Yaxiong

AU - Fu, Xianping

N1 - Funding: This work was supported in part by the National Key Research and Development Program of China Grant 2022YFE0132600, by the National Natural Science Foundation of China Grants 62002043, by the Research Project of China Disabled Persons’ Federation - on Assistive Technology Grant 2024CDPFAT-28.

PY - 2025/5/25

Y1 - 2025/5/25

N2 - To tackle the cross-domain model performance degradation challenge, a gaze estimation method based on proxy tuning is proposed, called PTGaze. In PTGaze, an base model is used to learn the gaze representation of the baseline model in the source domain, and an adapt model is used to learn the gaze representation of the baseline model in the target domain. The gaze difference between the base and adapt models is utilized to guide the final output, ensuring the method’s accuracy in the target domain. Experimental results show that the proposed method achieves higher cross-domain gaze estimation accuracy on five public datasets, using RT-Gene, Full-face, and Gaze360 as baseline models.

AB - To tackle the cross-domain model performance degradation challenge, a gaze estimation method based on proxy tuning is proposed, called PTGaze. In PTGaze, an base model is used to learn the gaze representation of the baseline model in the source domain, and an adapt model is used to learn the gaze representation of the baseline model in the target domain. The gaze difference between the base and adapt models is utilized to guide the final output, ensuring the method’s accuracy in the target domain. Experimental results show that the proposed method achieves higher cross-domain gaze estimation accuracy on five public datasets, using RT-Gene, Full-face, and Gaze360 as baseline models.

U2 - 10.1145/3715669.372682

DO - 10.1145/3715669.372682

M3 - Conference contribution

SN - 9798400714870

SP - 1

EP - 2

BT - Proceedings of the 2025 symposium on eye tracking research and applications (ETRA)

A2 - Sugano, Yusuke

A2 - Khamis, Mohamed

A2 - Chetouani, Aladine

A2 - Sidenmark, Ludwig

A2 - Bruno, Alessandro

PB - ACM

CY - New York

ER -

PTGaze: Cross-domain gaze estimation via proxy tuning

Abstract

Access to Document

Fingerprint

Cite this