Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation

Samitha Herath; Basura Fernando; Ehsan Abbasnejad; Munawar Hayat; Shahram Khadivi; Mehrtash Harandi; Hamid Rezatofighi; Gholamreza Haffari

doi:10.1109/ICCV51070.2023.01070

Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation

Samitha Herath, Basura Fernando, Ehsan Abbasnejad, Munawar Hayat, Shahram Khadivi, Mehrtash Harandi, Hamid Rezatofighi, Gholamreza Haffari

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

7 Citations (Scopus)

Abstract

We propose an Unsupervised Domain Adaptation (UDA) method by making use of Energy-Based Learning (EBL) and demonstrate 1. EBL can be used to improve the instance selection for a self-training task on the unlabelled target domain, and 2. alignment and normalizing energy scores can learn domain-invariant representations. For the former, we show that an energy-based selection criterion can be used to model instance selections by mimicking the joint distribution between data and predictions in the target domain. As per learning domain invariant representations, we show that stable domain alignment can be achieved by a combined energy alignment and an energy normalization process. We implement our method in consistent with the vision-transformer (ViT) backbone and show that our proposed method can outperform state-of-the-art ViT based UDA methods on diverse benchmarks (DomainNet, Office-Home, and VISDA2017).

Original language	English
Title of host publication	Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023
Editors	Frédéric Jurie, Gaurav Sharma
Place of Publication	Piscataway NJ USA
Publisher	IEEE, Institute of Electrical and Electronics Engineers
Pages	11619-11628
Number of pages	10
ISBN (Electronic)	9798350307184
ISBN (Print)	9798350307191
DOIs	https://doi.org/10.1109/ICCV51070.2023.01070
Publication status	Published - 2023
Event	IEEE International Conference on Computer Vision 2023 - Paris, France Duration: 2 Oct 2023 → 6 Oct 2023 https://ieeexplore.ieee.org/xpl/conhome/10376473/proceeding (Proceedings) https://iccv2023.thecvf.com/ (Website)

Conference

Conference	IEEE International Conference on Computer Vision 2023
Abbreviated title	ICCV 2023
Country/Territory	France
City	Paris
Period	2/10/23 → 6/10/23
Internet address	https://ieeexplore.ieee.org/xpl/conhome/10376473/proceeding (Proceedings) https://iccv2023.thecvf.com/ (Website)

Access to Document

10.1109/ICCV51070.2023.01070

Cite this

Herath, S., Fernando, B., Abbasnejad, E., Hayat, M., Khadivi, S., Harandi, M., Rezatofighi, H., & Haffari, G. (2023). Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation. In F. Jurie, & G. Sharma (Eds.), Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023 (pp. 11619-11628). IEEE, Institute of Electrical and Electronics Engineers. https://doi.org/10.1109/ICCV51070.2023.01070

Herath, Samitha ; Fernando, Basura ; Abbasnejad, Ehsan et al. / Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation. Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023. editor / Frédéric Jurie ; Gaurav Sharma. Piscataway NJ USA : IEEE, Institute of Electrical and Electronics Engineers, 2023. pp. 11619-11628

@inproceedings{d5b39c707b2b4b47b7059b3411ac1abe,

title = "Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation",

abstract = "We propose an Unsupervised Domain Adaptation (UDA) method by making use of Energy-Based Learning (EBL) and demonstrate 1. EBL can be used to improve the instance selection for a self-training task on the unlabelled target domain, and 2. alignment and normalizing energy scores can learn domain-invariant representations. For the former, we show that an energy-based selection criterion can be used to model instance selections by mimicking the joint distribution between data and predictions in the target domain. As per learning domain invariant representations, we show that stable domain alignment can be achieved by a combined energy alignment and an energy normalization process. We implement our method in consistent with the vision-transformer (ViT) backbone and show that our proposed method can outperform state-of-the-art ViT based UDA methods on diverse benchmarks (DomainNet, Office-Home, and VISDA2017).",

author = "Samitha Herath and Basura Fernando and Ehsan Abbasnejad and Munawar Hayat and Shahram Khadivi and Mehrtash Harandi and Hamid Rezatofighi and Gholamreza Haffari",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; IEEE International Conference on Computer Vision 2023, ICCV 2023 ; Conference date: 02-10-2023 Through 06-10-2023",

year = "2023",

doi = "10.1109/ICCV51070.2023.01070",

language = "English",

isbn = "9798350307191",

pages = "11619--11628",

editor = "Fr{\'e}d{\'e}ric Jurie and Gaurav Sharma",

booktitle = "Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023",

publisher = "IEEE, Institute of Electrical and Electronics Engineers",

address = "United States of America",

url = "https://ieeexplore.ieee.org/xpl/conhome/10376473/proceeding, https://iccv2023.thecvf.com/",

}

Herath, S, Fernando, B, Abbasnejad, E, Hayat, M, Khadivi, S, Harandi, M , Rezatofighi, H & Haffari, G 2023, Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation. in F Jurie & G Sharma (eds), Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023. IEEE, Institute of Electrical and Electronics Engineers, Piscataway NJ USA, pp. 11619-11628, IEEE International Conference on Computer Vision 2023, Paris, France, 2/10/23. https://doi.org/10.1109/ICCV51070.2023.01070

Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation. / Herath, Samitha; Fernando, Basura; Abbasnejad, Ehsan et al.
Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023. ed. / Frédéric Jurie; Gaurav Sharma. Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers, 2023. p. 11619-11628.

Research output: Chapter in Book/Report/Conference proceeding › Conference Paper › Research › peer-review

TY - GEN

T1 - Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation

AU - Herath, Samitha

AU - Fernando, Basura

AU - Abbasnejad, Ehsan

AU - Hayat, Munawar

AU - Khadivi, Shahram

AU - Harandi, Mehrtash

AU - Rezatofighi, Hamid

AU - Haffari, Gholamreza

PY - 2023

Y1 - 2023

N2 - We propose an Unsupervised Domain Adaptation (UDA) method by making use of Energy-Based Learning (EBL) and demonstrate 1. EBL can be used to improve the instance selection for a self-training task on the unlabelled target domain, and 2. alignment and normalizing energy scores can learn domain-invariant representations. For the former, we show that an energy-based selection criterion can be used to model instance selections by mimicking the joint distribution between data and predictions in the target domain. As per learning domain invariant representations, we show that stable domain alignment can be achieved by a combined energy alignment and an energy normalization process. We implement our method in consistent with the vision-transformer (ViT) backbone and show that our proposed method can outperform state-of-the-art ViT based UDA methods on diverse benchmarks (DomainNet, Office-Home, and VISDA2017).

AB - We propose an Unsupervised Domain Adaptation (UDA) method by making use of Energy-Based Learning (EBL) and demonstrate 1. EBL can be used to improve the instance selection for a self-training task on the unlabelled target domain, and 2. alignment and normalizing energy scores can learn domain-invariant representations. For the former, we show that an energy-based selection criterion can be used to model instance selections by mimicking the joint distribution between data and predictions in the target domain. As per learning domain invariant representations, we show that stable domain alignment can be achieved by a combined energy alignment and an energy normalization process. We implement our method in consistent with the vision-transformer (ViT) backbone and show that our proposed method can outperform state-of-the-art ViT based UDA methods on diverse benchmarks (DomainNet, Office-Home, and VISDA2017).

UR - http://www.scopus.com/inward/record.url?scp=85188242580&partnerID=8YFLogxK

U2 - 10.1109/ICCV51070.2023.01070

DO - 10.1109/ICCV51070.2023.01070

M3 - Conference Paper

AN - SCOPUS:85188242580

SN - 9798350307191

SP - 11619

EP - 11628

BT - Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023

A2 - Jurie, Frédéric

A2 - Sharma, Gaurav

PB - IEEE, Institute of Electrical and Electronics Engineers

CY - Piscataway NJ USA

T2 - IEEE International Conference on Computer Vision 2023

Y2 - 2 October 2023 through 6 October 2023

ER -

Herath S, Fernando B, Abbasnejad E, Hayat M, Khadivi S, Harandi M et al. Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation. In Jurie F, Sharma G, editors, Proceedings - 2023 IEEE/CVF International Conference on Computer Vision, ICCV 2023. Piscataway NJ USA: IEEE, Institute of Electrical and Electronics Engineers. 2023. p. 11619-11628 doi: 10.1109/ICCV51070.2023.01070

Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation

Abstract

Conference

Access to Document

Other files and links

Exploiting Geometries of Learning for Fast, Adaptive and Robust AI

Cite this

Energy-based Self-Training and Normalization for Unsupervised Domain Adaptation

Abstract

Conference

Access to Document

Other files and links

Projects

Exploiting Geometries of Learning for Fast, Adaptive and Robust AI

Cite this