TS 2 C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection

Yunchao Wei; Zhiqiang Shen; Bowen Cheng; Honghui Shi; Jinjun Xiong; Jiashi Feng; Thomas Huang

doi:10.1007/978-3-030-01252-6_27

TS ² C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection

Yunchao Wei, Zhiqiang Shen, Bowen Cheng, Honghui Shi, Jinjun Xiong, Jiashi Feng, Thomas Huang

Coordinated Science Lab

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

Abstract

This work provides a simple approach to discover tight object bounding boxes with only image-level supervision, called Tight box mining with Surrounding Segmentation Context (TS²C). We observe that object candidates mined through current multiple instance learning methods are usually trapped to discriminative object parts, rather than the entire object. TS²C leverages surrounding segmentation context derived from weakly-supervised segmentation to suppress such low-quality distracting candidates and boost the high-quality ones. Specifically, TS²C is developed based on two key properties of desirable bounding boxes: (1) high purity, meaning most pixels in the box are with high object response, and (2) high completeness, meaning the box covers high object response pixels comprehensively. With such novel and computable criteria, more tight candidates can be discovered for learning a better object detector. With TS²C, we obtain 48.0% and 44.4% mAP scores on VOC 2007 and 2012 benchmarks, which are the new state-of-the-arts.

Original language	English (US)
Title of host publication	Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
Editors	Vittorio Ferrari, Cristian Sminchisescu, Yair Weiss, Martial Hebert
Publisher	Springer
Pages	454-470
Number of pages	17
ISBN (Print)	9783030012519
DOIs	https://doi.org/10.1007/978-3-030-01252-6_27
State	Published - 2018
Event	15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany Duration: Sep 8 2018 → Sep 14 2018

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	11215 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Other

Other	15th European Conference on Computer Vision, ECCV 2018
Country/Territory	Germany
City	Munich
Period	9/8/18 → 9/14/18

Keywords

Object detection
Semantic segmentation
Weakly-supervised learning

ASJC Scopus subject areas

Theoretical Computer Science
General Computer Science

Online availability

10.1007/978-3-030-01252-6_27

Library availability

Discover UIUC Full Text

Cite this

Wei, Y., Shen, Z., Cheng, B., Shi, H., Xiong, J., Feng, J., & Huang, T. (2018). TS ² C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection. In V. Ferrari, C. Sminchisescu, Y. Weiss, & M. Hebert (Eds.), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings (pp. 454-470). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11215 LNCS). Springer. https://doi.org/10.1007/978-3-030-01252-6_27

TS ² C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection. / Wei, Yunchao; Shen, Zhiqiang; Cheng, Bowen et al.
Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. ed. / Vittorio Ferrari; Cristian Sminchisescu; Yair Weiss; Martial Hebert. Springer, 2018. p. 454-470 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11215 LNCS).

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

Wei, Y, Shen, Z, Cheng, B, Shi, H, Xiong, J, Feng, J & Huang, T 2018, TS ² C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection. in V Ferrari, C Sminchisescu, Y Weiss & M Hebert (eds), Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11215 LNCS, Springer, pp. 454-470, 15th European Conference on Computer Vision, ECCV 2018, Munich, Germany, 9/8/18. https://doi.org/10.1007/978-3-030-01252-6_27

Wei Y, Shen Z, Cheng B, Shi H, Xiong J, Feng J et al. TS ² C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection. In Ferrari V, Sminchisescu C, Weiss Y, Hebert M, editors, Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. Springer. 2018. p. 454-470. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-01252-6_27

Wei, Yunchao ; Shen, Zhiqiang ; Cheng, Bowen et al. / TS ² C : Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection. Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. editor / Vittorio Ferrari ; Cristian Sminchisescu ; Yair Weiss ; Martial Hebert. Springer, 2018. pp. 454-470 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{0d9b99d5fd3340ae8eabd1504a845e83,

title = "TS 2 C: Tight Box Mining with Surrounding Segmentation Context for Weakly Supervised Object Detection",

abstract = "This work provides a simple approach to discover tight object bounding boxes with only image-level supervision, called Tight box mining with Surrounding Segmentation Context (TS2C). We observe that object candidates mined through current multiple instance learning methods are usually trapped to discriminative object parts, rather than the entire object. TS2C leverages surrounding segmentation context derived from weakly-supervised segmentation to suppress such low-quality distracting candidates and boost the high-quality ones. Specifically, TS2C is developed based on two key properties of desirable bounding boxes: (1) high purity, meaning most pixels in the box are with high object response, and (2) high completeness, meaning the box covers high object response pixels comprehensively. With such novel and computable criteria, more tight candidates can be discovered for learning a better object detector. With TS2C, we obtain 48.0% and 44.4% mAP scores on VOC 2007 and 2012 benchmarks, which are the new state-of-the-arts.",

keywords = "Object detection, Semantic segmentation, Weakly-supervised learning",

author = "Yunchao Wei and Zhiqiang Shen and Bowen Cheng and Honghui Shi and Jinjun Xiong and Jiashi Feng and Thomas Huang",

note = "Publisher Copyright: {\textcopyright} 2018, Springer Nature Switzerland AG.; 15th European Conference on Computer Vision, ECCV 2018 ; Conference date: 08-09-2018 Through 14-09-2018",

year = "2018",

doi = "10.1007/978-3-030-01252-6_27",

language = "English (US)",

isbn = "9783030012519",

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

publisher = "Springer",

pages = "454--470",

editor = "Vittorio Ferrari and Cristian Sminchisescu and Yair Weiss and Martial Hebert",

booktitle = "Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings",

address = "Germany",

}

TY - GEN

T1 - TS 2 C

T2 - 15th European Conference on Computer Vision, ECCV 2018

AU - Wei, Yunchao

AU - Shen, Zhiqiang

AU - Cheng, Bowen

AU - Shi, Honghui

AU - Xiong, Jinjun

AU - Feng, Jiashi

AU - Huang, Thomas

PY - 2018

Y1 - 2018

N2 - This work provides a simple approach to discover tight object bounding boxes with only image-level supervision, called Tight box mining with Surrounding Segmentation Context (TS2C). We observe that object candidates mined through current multiple instance learning methods are usually trapped to discriminative object parts, rather than the entire object. TS2C leverages surrounding segmentation context derived from weakly-supervised segmentation to suppress such low-quality distracting candidates and boost the high-quality ones. Specifically, TS2C is developed based on two key properties of desirable bounding boxes: (1) high purity, meaning most pixels in the box are with high object response, and (2) high completeness, meaning the box covers high object response pixels comprehensively. With such novel and computable criteria, more tight candidates can be discovered for learning a better object detector. With TS2C, we obtain 48.0% and 44.4% mAP scores on VOC 2007 and 2012 benchmarks, which are the new state-of-the-arts.

AB - This work provides a simple approach to discover tight object bounding boxes with only image-level supervision, called Tight box mining with Surrounding Segmentation Context (TS2C). We observe that object candidates mined through current multiple instance learning methods are usually trapped to discriminative object parts, rather than the entire object. TS2C leverages surrounding segmentation context derived from weakly-supervised segmentation to suppress such low-quality distracting candidates and boost the high-quality ones. Specifically, TS2C is developed based on two key properties of desirable bounding boxes: (1) high purity, meaning most pixels in the box are with high object response, and (2) high completeness, meaning the box covers high object response pixels comprehensively. With such novel and computable criteria, more tight candidates can be discovered for learning a better object detector. With TS2C, we obtain 48.0% and 44.4% mAP scores on VOC 2007 and 2012 benchmarks, which are the new state-of-the-arts.

KW - Object detection

KW - Semantic segmentation

KW - Weakly-supervised learning

UR - http://www.scopus.com/inward/record.url?scp=85055106295&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85055106295&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-01252-6_27

DO - 10.1007/978-3-030-01252-6_27

M3 - Conference contribution

AN - SCOPUS:85055106295

SN - 9783030012519

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

SP - 454

EP - 470

BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings

A2 - Ferrari, Vittorio

A2 - Sminchisescu, Cristian

A2 - Weiss, Yair

A2 - Hebert, Martial

PB - Springer

Y2 - 8 September 2018 through 14 September 2018

ER -