Attentional feature fusion

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Standard

Attentional feature fusion. / Dai, Yimian; Gieseke, Fabian; Oehmcke, Stefan; Wu, Yiquan; Barnard, Kobus.

Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021. IEEE, 2021. s. 3559-3568.

Publikation: Bidrag til bog/antologi/rapportKonferencebidrag i proceedingsForskningfagfællebedømt

Harvard

Dai, Y, Gieseke, F, Oehmcke, S, Wu, Y & Barnard, K 2021, Attentional feature fusion. i Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021. IEEE, s. 3559-3568, 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021, Virtual, Online, USA, 05/01/2021. https://doi.org/10.1109/WACV48630.2021.00360

APA

Dai, Y., Gieseke, F., Oehmcke, S., Wu, Y., & Barnard, K. (2021). Attentional feature fusion. I Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021 (s. 3559-3568). IEEE. https://doi.org/10.1109/WACV48630.2021.00360

Vancouver

Dai Y, Gieseke F, Oehmcke S, Wu Y, Barnard K. Attentional feature fusion. I Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021. IEEE. 2021. s. 3559-3568 https://doi.org/10.1109/WACV48630.2021.00360

Author

Dai, Yimian ; Gieseke, Fabian ; Oehmcke, Stefan ; Wu, Yiquan ; Barnard, Kobus. / Attentional feature fusion. Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021. IEEE, 2021. s. 3559-3568

Bibtex

@inproceedings{197a8ca5f02a4bc88e13c501e3aab1bd,
title = "Attentional feature fusion",
abstract = "Feature fusion, the combination of features from different layers or branches, is an omnipresent part of modern network architectures. It is often implemented via simple operations, such as summation or concatenation, but this might not be the best choice. In this work, we propose a uniform and general scheme, namely attentional feature fusion, which is applicable for most common scenarios, including feature fusion induced by short and long skip connections as well as within Inception layers. To better fuse features of inconsistent semantics and scales, we propose a multiscale channel attention module, which addresses issues that arise when fusing features given at different scales. We also demonstrate that the initial integration of feature maps can become a bottleneck and that this issue can be alleviated by adding another level of attention, which we refer to as iterative attentional feature fusion. With fewer layers or parameters, our models outperform state-of-the-art networks on both CIFAR-100 and ImageNet datasets, which suggests that more sophisticated attention mechanisms for feature fusion hold great potential to consistently yield better results compared to their direct counterparts. Our codes and trained models are available online1.",
author = "Yimian Dai and Fabian Gieseke and Stefan Oehmcke and Yiquan Wu and Kobus Barnard",
note = "Publisher Copyright: {\textcopyright} 2021 IEEE.; 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021 ; Conference date: 05-01-2021 Through 09-01-2021",
year = "2021",
doi = "10.1109/WACV48630.2021.00360",
language = "English",
pages = "3559--3568",
booktitle = "Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021",
publisher = "IEEE",

}

RIS

TY - GEN

T1 - Attentional feature fusion

AU - Dai, Yimian

AU - Gieseke, Fabian

AU - Oehmcke, Stefan

AU - Wu, Yiquan

AU - Barnard, Kobus

N1 - Publisher Copyright: © 2021 IEEE.

PY - 2021

Y1 - 2021

N2 - Feature fusion, the combination of features from different layers or branches, is an omnipresent part of modern network architectures. It is often implemented via simple operations, such as summation or concatenation, but this might not be the best choice. In this work, we propose a uniform and general scheme, namely attentional feature fusion, which is applicable for most common scenarios, including feature fusion induced by short and long skip connections as well as within Inception layers. To better fuse features of inconsistent semantics and scales, we propose a multiscale channel attention module, which addresses issues that arise when fusing features given at different scales. We also demonstrate that the initial integration of feature maps can become a bottleneck and that this issue can be alleviated by adding another level of attention, which we refer to as iterative attentional feature fusion. With fewer layers or parameters, our models outperform state-of-the-art networks on both CIFAR-100 and ImageNet datasets, which suggests that more sophisticated attention mechanisms for feature fusion hold great potential to consistently yield better results compared to their direct counterparts. Our codes and trained models are available online1.

AB - Feature fusion, the combination of features from different layers or branches, is an omnipresent part of modern network architectures. It is often implemented via simple operations, such as summation or concatenation, but this might not be the best choice. In this work, we propose a uniform and general scheme, namely attentional feature fusion, which is applicable for most common scenarios, including feature fusion induced by short and long skip connections as well as within Inception layers. To better fuse features of inconsistent semantics and scales, we propose a multiscale channel attention module, which addresses issues that arise when fusing features given at different scales. We also demonstrate that the initial integration of feature maps can become a bottleneck and that this issue can be alleviated by adding another level of attention, which we refer to as iterative attentional feature fusion. With fewer layers or parameters, our models outperform state-of-the-art networks on both CIFAR-100 and ImageNet datasets, which suggests that more sophisticated attention mechanisms for feature fusion hold great potential to consistently yield better results compared to their direct counterparts. Our codes and trained models are available online1.

UR - http://www.scopus.com/inward/record.url?scp=85113484606&partnerID=8YFLogxK

U2 - 10.1109/WACV48630.2021.00360

DO - 10.1109/WACV48630.2021.00360

M3 - Article in proceedings

AN - SCOPUS:85113484606

SP - 3559

EP - 3568

BT - Proceedings - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021

PB - IEEE

T2 - 2021 IEEE Winter Conference on Applications of Computer Vision, WACV 2021

Y2 - 5 January 2021 through 9 January 2021

ER -

ID: 282740108