Long-Tailed Instance Segmentation Using Gumbel Optimized Loss

Kostas Alexandridis; Jiankang Deng; Anh Nguyen; Shan Luo

doi:10.1007/978-3-031-20080-9_21

Long-Tailed Instance Segmentation Using Gumbel Optimized Loss

Kostas Alexandridis, Jiankang Deng, Anh Nguyen, Shan Luo^*

^*Corresponding author for this work

Engineering

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

17 Citations (Scopus)

Abstract

Major advancements have been made in the field of object detection and segmentation recently. However, when it comes to rare categories, the state-of-the-art methods fail to detect them, resulting in a significant performance gap between rare and frequent categories. In this paper, we identify that Sigmoid or Softmax functions used in deep detectors are a major reason for low performance and are sub-optimal for long-tailed detection and segmentation. To address this, we develop a Gumbel Optimized Loss (GOL), for long-tailed detection and segmentation. It aligns with the Gumbel distribution of rare classes in imbalanced datasets, considering the fact that most classes in long-tailed detection have low expected probability. The proposed GOL significantly outperforms the best state-of-the-art method by 1.1 % on AP, and boosts the overall segmentation by 9.0 % and detection by 8.0 %, particularly improving detection of rare classes by 20.3 %, compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.

Original language	English
Title of host publication	Computer Vision – ECCV 2022 - 17th European Conference, Proceedings
Editors	Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, Tal Hassner
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	353-369
Number of pages	17
ISBN (Print)	9783031200793
DOIs	https://doi.org/10.1007/978-3-031-20080-9_21
Publication status	Published - 2022
Event	17th European Conference on Computer Vision, ECCV 2022 - Tel Aviv, Israel Duration: 23 Oct 2022 → 27 Oct 2022

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	13670 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	17th European Conference on Computer Vision, ECCV 2022
Country/Territory	Israel
City	Tel Aviv
Period	23/10/2022 → 27/10/2022

Keywords

Gumbel activation
Long-tailed distribution
Long-tailed instance segmentation

Access to Document

10.1007/978-3-031-20080-9_21

Cite this

Alexandridis, K., Deng, J., Nguyen, A., & Luo, S. (2022). Long-Tailed Instance Segmentation Using Gumbel Optimized Loss. In S. Avidan, G. Brostow, M. Cissé, G. M. Farinella, & T. Hassner (Eds.), Computer Vision – ECCV 2022 - 17th European Conference, Proceedings (pp. 353-369). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13670 LNCS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-031-20080-9_21

Alexandridis, Kostas ; Deng, Jiankang ; Nguyen, Anh et al. / Long-Tailed Instance Segmentation Using Gumbel Optimized Loss. Computer Vision – ECCV 2022 - 17th European Conference, Proceedings. editor / Shai Avidan ; Gabriel Brostow ; Moustapha Cissé ; Giovanni Maria Farinella ; Tal Hassner. Springer Science and Business Media Deutschland GmbH, 2022. pp. 353-369 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inbook{dc4e91b56a084915a43a885a7cdd1a43,

title = "Long-Tailed Instance Segmentation Using Gumbel Optimized Loss",

abstract = "Major advancements have been made in the field of object detection and segmentation recently. However, when it comes to rare categories, the state-of-the-art methods fail to detect them, resulting in a significant performance gap between rare and frequent categories. In this paper, we identify that Sigmoid or Softmax functions used in deep detectors are a major reason for low performance and are sub-optimal for long-tailed detection and segmentation. To address this, we develop a Gumbel Optimized Loss (GOL), for long-tailed detection and segmentation. It aligns with the Gumbel distribution of rare classes in imbalanced datasets, considering the fact that most classes in long-tailed detection have low expected probability. The proposed GOL significantly outperforms the best state-of-the-art method by 1.1 % on AP, and boosts the overall segmentation by 9.0 % and detection by 8.0 %, particularly improving detection of rare classes by 20.3 %, compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.",

keywords = "Gumbel activation, Long-tailed distribution, Long-tailed instance segmentation",

author = "Kostas Alexandridis and Jiankang Deng and Anh Nguyen and Shan Luo",

note = "Funding Information: Acknowledgments. This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) Centre for Doctoral Training in Distributed Algorithms [EP/S023445/1]; EPSRC ViTac project (EP/T033517/1); King{\textquoteright}s College London NMESFS PhD Studentship; the University of Liverpool and Vision4ce. It also made use of the facilities of the N8 Centre of Excellence in Computationally Intensive Research provided and funded by the N8 research partnership and EPSRC [EP/T022167/1]. Publisher Copyright: {\textcopyright} 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.; 17th European Conference on Computer Vision, ECCV 2022 ; Conference date: 23-10-2022 Through 27-10-2022",

year = "2022",

doi = "10.1007/978-3-031-20080-9_21",

language = "English",

isbn = "9783031200793",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "353--369",

editor = "Shai Avidan and Gabriel Brostow and Moustapha Ciss{\'e} and Farinella, {Giovanni Maria} and Tal Hassner",

booktitle = "Computer Vision – ECCV 2022 - 17th European Conference, Proceedings",

address = "Germany",

}

Alexandridis, K, Deng, J, Nguyen, A & Luo, S 2022, Long-Tailed Instance Segmentation Using Gumbel Optimized Loss. in S Avidan, G Brostow, M Cissé, GM Farinella & T Hassner (eds), Computer Vision – ECCV 2022 - 17th European Conference, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13670 LNCS, Springer Science and Business Media Deutschland GmbH, pp. 353-369, 17th European Conference on Computer Vision, ECCV 2022, Tel Aviv, Israel, 23/10/2022. https://doi.org/10.1007/978-3-031-20080-9_21

Long-Tailed Instance Segmentation Using Gumbel Optimized Loss. / Alexandridis, Kostas; Deng, Jiankang; Nguyen, Anh et al.
Computer Vision – ECCV 2022 - 17th European Conference, Proceedings. ed. / Shai Avidan; Gabriel Brostow; Moustapha Cissé; Giovanni Maria Farinella; Tal Hassner. Springer Science and Business Media Deutschland GmbH, 2022. p. 353-369 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 13670 LNCS).

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

TY - CHAP

T1 - Long-Tailed Instance Segmentation Using Gumbel Optimized Loss

AU - Alexandridis, Kostas

AU - Deng, Jiankang

AU - Nguyen, Anh

AU - Luo, Shan

N1 - Funding Information: Acknowledgments. This work was supported by the Engineering and Physical Sciences Research Council (EPSRC) Centre for Doctoral Training in Distributed Algorithms [EP/S023445/1]; EPSRC ViTac project (EP/T033517/1); King’s College London NMESFS PhD Studentship; the University of Liverpool and Vision4ce. It also made use of the facilities of the N8 Centre of Excellence in Computationally Intensive Research provided and funded by the N8 research partnership and EPSRC [EP/T022167/1]. Publisher Copyright: © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

PY - 2022

Y1 - 2022

N2 - Major advancements have been made in the field of object detection and segmentation recently. However, when it comes to rare categories, the state-of-the-art methods fail to detect them, resulting in a significant performance gap between rare and frequent categories. In this paper, we identify that Sigmoid or Softmax functions used in deep detectors are a major reason for low performance and are sub-optimal for long-tailed detection and segmentation. To address this, we develop a Gumbel Optimized Loss (GOL), for long-tailed detection and segmentation. It aligns with the Gumbel distribution of rare classes in imbalanced datasets, considering the fact that most classes in long-tailed detection have low expected probability. The proposed GOL significantly outperforms the best state-of-the-art method by 1.1 % on AP, and boosts the overall segmentation by 9.0 % and detection by 8.0 %, particularly improving detection of rare classes by 20.3 %, compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.

AB - Major advancements have been made in the field of object detection and segmentation recently. However, when it comes to rare categories, the state-of-the-art methods fail to detect them, resulting in a significant performance gap between rare and frequent categories. In this paper, we identify that Sigmoid or Softmax functions used in deep detectors are a major reason for low performance and are sub-optimal for long-tailed detection and segmentation. To address this, we develop a Gumbel Optimized Loss (GOL), for long-tailed detection and segmentation. It aligns with the Gumbel distribution of rare classes in imbalanced datasets, considering the fact that most classes in long-tailed detection have low expected probability. The proposed GOL significantly outperforms the best state-of-the-art method by 1.1 % on AP, and boosts the overall segmentation by 9.0 % and detection by 8.0 %, particularly improving detection of rare classes by 20.3 %, compared to Mask-RCNN, on LVIS dataset. Code available at: https://github.com/kostas1515/GOL.

KW - Gumbel activation

KW - Long-tailed distribution

KW - Long-tailed instance segmentation

UR - http://www.scopus.com/inward/record.url?scp=85144554945&partnerID=8YFLogxK

U2 - 10.1007/978-3-031-20080-9_21

DO - 10.1007/978-3-031-20080-9_21

M3 - Conference paper

AN - SCOPUS:85144554945

SN - 9783031200793

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 353

EP - 369

BT - Computer Vision – ECCV 2022 - 17th European Conference, Proceedings

A2 - Avidan, Shai

A2 - Brostow, Gabriel

A2 - Cissé, Moustapha

A2 - Farinella, Giovanni Maria

A2 - Hassner, Tal

PB - Springer Science and Business Media Deutschland GmbH

T2 - 17th European Conference on Computer Vision, ECCV 2022

Y2 - 23 October 2022 through 27 October 2022

ER -

Alexandridis K, Deng J, Nguyen A, Luo S. Long-Tailed Instance Segmentation Using Gumbel Optimized Loss. In Avidan S, Brostow G, Cissé M, Farinella GM, Hassner T, editors, Computer Vision – ECCV 2022 - 17th European Conference, Proceedings. Springer Science and Business Media Deutschland GmbH. 2022. p. 353-369. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-031-20080-9_21

Long-Tailed Instance Segmentation Using Gumbel Optimized Loss

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this