IEICE TRANSACTIONS on Information
Model-Agnostic Multi-Domain Learning with Domain-Specific Adapters for Action Recognition
Summary :
In this paper, we propose a multi-domain learning model for action recognition. The proposed method inserts domain-specific adapters between layers of domain-independent layers of a backbone network. Unlike a multi-head network that switches classification heads only, our model switches not only the heads, but also the adapters for facilitating to learn feature representations universal to multiple domains. Unlike prior works, the proposed method is model-agnostic and doesn't assume model structures unlike prior works. Experimental results on three popular action recognition datasets (HMDB51, UCF101, and Kinetics-400) demonstrate that the proposed method is more effective than a multi-head architecture and more efficient than separately training models for each domain.
- Publication
- IEICE TRANSACTIONS on Information Vol.E105-D No.12 pp.2119-2126
- Publication Date
- 2022/12/01
- Publicized
- 2022/09/15
- Online ISSN
- 1745-1361
- DOI
- 10.1587/transinf.2022EDP7058
- Type of Manuscript
- PAPER
- Category
- Image Recognition, Computer Vision
Authors
Kazuki OMI
Nagoya Institute of Technology
Jun KIMATA
Nagoya Institute of Technology
Toru TAMAKI
Nagoya Institute of Technology
Keyword
Latest Issue
Copyrights notice of machine-translated contents
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Cite this
Copy
Kazuki OMI, Jun KIMATA, Toru TAMAKI, "Model-Agnostic Multi-Domain Learning with Domain-Specific Adapters for Action Recognition" in IEICE TRANSACTIONS on Information,
vol. E105-D, no. 12, pp. 2119-2126, December 2022, doi: 10.1587/transinf.2022EDP7058.
Abstract: In this paper, we propose a multi-domain learning model for action recognition. The proposed method inserts domain-specific adapters between layers of domain-independent layers of a backbone network. Unlike a multi-head network that switches classification heads only, our model switches not only the heads, but also the adapters for facilitating to learn feature representations universal to multiple domains. Unlike prior works, the proposed method is model-agnostic and doesn't assume model structures unlike prior works. Experimental results on three popular action recognition datasets (HMDB51, UCF101, and Kinetics-400) demonstrate that the proposed method is more effective than a multi-head architecture and more efficient than separately training models for each domain.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2022EDP7058/_p
Copy
@ARTICLE{e105-d_12_2119,
author={Kazuki OMI, Jun KIMATA, Toru TAMAKI, },
journal={IEICE TRANSACTIONS on Information},
title={Model-Agnostic Multi-Domain Learning with Domain-Specific Adapters for Action Recognition},
year={2022},
volume={E105-D},
number={12},
pages={2119-2126},
abstract={In this paper, we propose a multi-domain learning model for action recognition. The proposed method inserts domain-specific adapters between layers of domain-independent layers of a backbone network. Unlike a multi-head network that switches classification heads only, our model switches not only the heads, but also the adapters for facilitating to learn feature representations universal to multiple domains. Unlike prior works, the proposed method is model-agnostic and doesn't assume model structures unlike prior works. Experimental results on three popular action recognition datasets (HMDB51, UCF101, and Kinetics-400) demonstrate that the proposed method is more effective than a multi-head architecture and more efficient than separately training models for each domain.},
keywords={},
doi={10.1587/transinf.2022EDP7058},
ISSN={1745-1361},
month={December},}
Copy
TY - JOUR
TI - Model-Agnostic Multi-Domain Learning with Domain-Specific Adapters for Action Recognition
T2 - IEICE TRANSACTIONS on Information
SP - 2119
EP - 2126
AU - Kazuki OMI
AU - Jun KIMATA
AU - Toru TAMAKI
PY - 2022
DO - 10.1587/transinf.2022EDP7058
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E105-D
IS - 12
JA - IEICE TRANSACTIONS on Information
Y1 - December 2022
AB - In this paper, we propose a multi-domain learning model for action recognition. The proposed method inserts domain-specific adapters between layers of domain-independent layers of a backbone network. Unlike a multi-head network that switches classification heads only, our model switches not only the heads, but also the adapters for facilitating to learn feature representations universal to multiple domains. Unlike prior works, the proposed method is model-agnostic and doesn't assume model structures unlike prior works. Experimental results on three popular action recognition datasets (HMDB51, UCF101, and Kinetics-400) demonstrate that the proposed method is more effective than a multi-head architecture and more efficient than separately training models for each domain.
ER -
English
