Deep Semantic Matching with Foreground Detection and Cycle-Consistency

Yun Chun Chen, Po Hsiang Huang, Li Yu Yu, Jia Bin Huang, Ming Hsuan Yang, Yen Yu Lin

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Establishing dense semantic correspondences between object instances remains a challenging problem due to background clutter, significant scale and pose differences, and large intra-class variations. In this paper, we present an end-to-end trainable network for learning semantic correspondences using only matching image pairs without manual keypoint correspondence annotations. To facilitate network training with this weaker form of supervision, we (1) explicitly estimate the foreground regions to suppress the effect of background clutter and (2) develop cycle-consistent losses to enforce the predicted transformations across multiple images to be geometrically plausible and consistent. We train the proposed model using the PF-PASCAL dataset and evaluate the performance on the PF-PASCAL, PF-WILLOW, and TSS datasets. Extensive experimental results show that the proposed approach achieves favorably performance compared to the state-of-the-art. The code and model will be available at https://yunchunchen.github.io/WeakMatchNet/.

Original languageEnglish
Title of host publicationComputer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers
EditorsC.V. Jawahar, Konrad Schindler, Hongdong Li, Greg Mori
PublisherSpringer Verlag
Pages347-362
Number of pages16
ISBN (Print)9783030208929
DOIs
Publication statusPublished - 2019 Jan 1
Event14th Asian Conference on Computer Vision, ACCV 2018 - Perth, Australia
Duration: 2018 Dec 22018 Dec 6

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume11363 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference14th Asian Conference on Computer Vision, ACCV 2018
CountryAustralia
CityPerth
Period18/12/218/12/6

Fingerprint

Correspondence
Semantics
Clutter
Cycle
Image matching
Image Matching
Annotation
Evaluate
Experimental Results
Model
Estimate
Background
Training
Object
Class
Form
Learning

All Science Journal Classification (ASJC) codes

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Chen, Y. C., Huang, P. H., Yu, L. Y., Huang, J. B., Yang, M. H., & Lin, Y. Y. (2019). Deep Semantic Matching with Foreground Detection and Cycle-Consistency. In C. V. Jawahar, K. Schindler, H. Li, & G. Mori (Eds.), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers (pp. 347-362). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11363 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-20893-6_22
Chen, Yun Chun ; Huang, Po Hsiang ; Yu, Li Yu ; Huang, Jia Bin ; Yang, Ming Hsuan ; Lin, Yen Yu. / Deep Semantic Matching with Foreground Detection and Cycle-Consistency. Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. editor / C.V. Jawahar ; Konrad Schindler ; Hongdong Li ; Greg Mori. Springer Verlag, 2019. pp. 347-362 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{7360c16f4b5e4dbf890aa7bd6c042c70,
title = "Deep Semantic Matching with Foreground Detection and Cycle-Consistency",
abstract = "Establishing dense semantic correspondences between object instances remains a challenging problem due to background clutter, significant scale and pose differences, and large intra-class variations. In this paper, we present an end-to-end trainable network for learning semantic correspondences using only matching image pairs without manual keypoint correspondence annotations. To facilitate network training with this weaker form of supervision, we (1) explicitly estimate the foreground regions to suppress the effect of background clutter and (2) develop cycle-consistent losses to enforce the predicted transformations across multiple images to be geometrically plausible and consistent. We train the proposed model using the PF-PASCAL dataset and evaluate the performance on the PF-PASCAL, PF-WILLOW, and TSS datasets. Extensive experimental results show that the proposed approach achieves favorably performance compared to the state-of-the-art. The code and model will be available at https://yunchunchen.github.io/WeakMatchNet/.",
author = "Chen, {Yun Chun} and Huang, {Po Hsiang} and Yu, {Li Yu} and Huang, {Jia Bin} and Yang, {Ming Hsuan} and Lin, {Yen Yu}",
year = "2019",
month = "1",
day = "1",
doi = "10.1007/978-3-030-20893-6_22",
language = "English",
isbn = "9783030208929",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "347--362",
editor = "C.V. Jawahar and Konrad Schindler and Hongdong Li and Greg Mori",
booktitle = "Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers",
address = "Germany",

}

Chen, YC, Huang, PH, Yu, LY, Huang, JB, Yang, MH & Lin, YY 2019, Deep Semantic Matching with Foreground Detection and Cycle-Consistency. in CV Jawahar, K Schindler, H Li & G Mori (eds), Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 11363 LNCS, Springer Verlag, pp. 347-362, 14th Asian Conference on Computer Vision, ACCV 2018, Perth, Australia, 18/12/2. https://doi.org/10.1007/978-3-030-20893-6_22

Deep Semantic Matching with Foreground Detection and Cycle-Consistency. / Chen, Yun Chun; Huang, Po Hsiang; Yu, Li Yu; Huang, Jia Bin; Yang, Ming Hsuan; Lin, Yen Yu.

Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. ed. / C.V. Jawahar; Konrad Schindler; Hongdong Li; Greg Mori. Springer Verlag, 2019. p. 347-362 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11363 LNCS).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

TY - GEN

T1 - Deep Semantic Matching with Foreground Detection and Cycle-Consistency

AU - Chen, Yun Chun

AU - Huang, Po Hsiang

AU - Yu, Li Yu

AU - Huang, Jia Bin

AU - Yang, Ming Hsuan

AU - Lin, Yen Yu

PY - 2019/1/1

Y1 - 2019/1/1

N2 - Establishing dense semantic correspondences between object instances remains a challenging problem due to background clutter, significant scale and pose differences, and large intra-class variations. In this paper, we present an end-to-end trainable network for learning semantic correspondences using only matching image pairs without manual keypoint correspondence annotations. To facilitate network training with this weaker form of supervision, we (1) explicitly estimate the foreground regions to suppress the effect of background clutter and (2) develop cycle-consistent losses to enforce the predicted transformations across multiple images to be geometrically plausible and consistent. We train the proposed model using the PF-PASCAL dataset and evaluate the performance on the PF-PASCAL, PF-WILLOW, and TSS datasets. Extensive experimental results show that the proposed approach achieves favorably performance compared to the state-of-the-art. The code and model will be available at https://yunchunchen.github.io/WeakMatchNet/.

AB - Establishing dense semantic correspondences between object instances remains a challenging problem due to background clutter, significant scale and pose differences, and large intra-class variations. In this paper, we present an end-to-end trainable network for learning semantic correspondences using only matching image pairs without manual keypoint correspondence annotations. To facilitate network training with this weaker form of supervision, we (1) explicitly estimate the foreground regions to suppress the effect of background clutter and (2) develop cycle-consistent losses to enforce the predicted transformations across multiple images to be geometrically plausible and consistent. We train the proposed model using the PF-PASCAL dataset and evaluate the performance on the PF-PASCAL, PF-WILLOW, and TSS datasets. Extensive experimental results show that the proposed approach achieves favorably performance compared to the state-of-the-art. The code and model will be available at https://yunchunchen.github.io/WeakMatchNet/.

UR - http://www.scopus.com/inward/record.url?scp=85067235030&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85067235030&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-20893-6_22

DO - 10.1007/978-3-030-20893-6_22

M3 - Conference contribution

AN - SCOPUS:85067235030

SN - 9783030208929

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

SP - 347

EP - 362

BT - Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers

A2 - Jawahar, C.V.

A2 - Schindler, Konrad

A2 - Li, Hongdong

A2 - Mori, Greg

PB - Springer Verlag

ER -

Chen YC, Huang PH, Yu LY, Huang JB, Yang MH, Lin YY. Deep Semantic Matching with Foreground Detection and Cycle-Consistency. In Jawahar CV, Schindler K, Li H, Mori G, editors, Computer Vision - ACCV 2018 - 14th Asian Conference on Computer Vision, Revised Selected Papers. Springer Verlag. 2019. p. 347-362. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-030-20893-6_22