Unsupervised stereo matching using correspondence consistency

Sunghun Joung, Seungryong Kim, Bumsub Ham, Kwanghoon Sohn

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

1 Citation (Scopus)

Abstract

Deep convolutional neural networks (CNNs) have shown revolutionary performance improvements for matching cost computation in stereo matching. However, conventional CNN-based approaches to learn the network in a supervised manner require a large number of ground-truth disparity maps, which limits their applicability. To overcome this limitation, we present a novel framework to learn a CNNs architecture for matching cost computation in an unsupervised manner. Our method leverages an image domain learning combined with stereo epipolar constraints. Exploiting the correspondence consistency between stereo images as supervision, our method selects the training samples in each iteration during network training and uses them to learn the network. To boost the performance, we also propose a multi-scale cost computation scheme. Experimental results show that our method outperforms the state-of-the-art methods including even supervised learning based methods on various benchmarks.

Original languageEnglish
Title of host publication2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings
PublisherIEEE Computer Society
Pages2518-2522
Number of pages5
ISBN (Electronic)9781509021758
DOIs
Publication statusPublished - 2018 Feb 20
Event24th IEEE International Conference on Image Processing, ICIP 2017 - Beijing, China
Duration: 2017 Sep 172017 Sep 20

Publication series

NameProceedings - International Conference on Image Processing, ICIP
Volume2017-September
ISSN (Print)1522-4880

Other

Other24th IEEE International Conference on Image Processing, ICIP 2017
CountryChina
CityBeijing
Period17/9/1717/9/20

Fingerprint

Neural networks
Costs
Supervised learning
Network architecture

All Science Journal Classification (ASJC) codes

  • Software
  • Computer Vision and Pattern Recognition
  • Signal Processing

Cite this

Joung, S., Kim, S., Ham, B., & Sohn, K. (2018). Unsupervised stereo matching using correspondence consistency. In 2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings (pp. 2518-2522). (Proceedings - International Conference on Image Processing, ICIP; Vol. 2017-September). IEEE Computer Society. https://doi.org/10.1109/ICIP.2017.8296736
Joung, Sunghun ; Kim, Seungryong ; Ham, Bumsub ; Sohn, Kwanghoon. / Unsupervised stereo matching using correspondence consistency. 2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings. IEEE Computer Society, 2018. pp. 2518-2522 (Proceedings - International Conference on Image Processing, ICIP).
@inproceedings{204aa75de0c34650b34d7bfb7699d7eb,
title = "Unsupervised stereo matching using correspondence consistency",
abstract = "Deep convolutional neural networks (CNNs) have shown revolutionary performance improvements for matching cost computation in stereo matching. However, conventional CNN-based approaches to learn the network in a supervised manner require a large number of ground-truth disparity maps, which limits their applicability. To overcome this limitation, we present a novel framework to learn a CNNs architecture for matching cost computation in an unsupervised manner. Our method leverages an image domain learning combined with stereo epipolar constraints. Exploiting the correspondence consistency between stereo images as supervision, our method selects the training samples in each iteration during network training and uses them to learn the network. To boost the performance, we also propose a multi-scale cost computation scheme. Experimental results show that our method outperforms the state-of-the-art methods including even supervised learning based methods on various benchmarks.",
author = "Sunghun Joung and Seungryong Kim and Bumsub Ham and Kwanghoon Sohn",
year = "2018",
month = "2",
day = "20",
doi = "10.1109/ICIP.2017.8296736",
language = "English",
series = "Proceedings - International Conference on Image Processing, ICIP",
publisher = "IEEE Computer Society",
pages = "2518--2522",
booktitle = "2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings",
address = "United States",

}

Joung, S, Kim, S, Ham, B & Sohn, K 2018, Unsupervised stereo matching using correspondence consistency. in 2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings. Proceedings - International Conference on Image Processing, ICIP, vol. 2017-September, IEEE Computer Society, pp. 2518-2522, 24th IEEE International Conference on Image Processing, ICIP 2017, Beijing, China, 17/9/17. https://doi.org/10.1109/ICIP.2017.8296736

Unsupervised stereo matching using correspondence consistency. / Joung, Sunghun; Kim, Seungryong; Ham, Bumsub; Sohn, Kwanghoon.

2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings. IEEE Computer Society, 2018. p. 2518-2522 (Proceedings - International Conference on Image Processing, ICIP; Vol. 2017-September).

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

TY - GEN

T1 - Unsupervised stereo matching using correspondence consistency

AU - Joung, Sunghun

AU - Kim, Seungryong

AU - Ham, Bumsub

AU - Sohn, Kwanghoon

PY - 2018/2/20

Y1 - 2018/2/20

N2 - Deep convolutional neural networks (CNNs) have shown revolutionary performance improvements for matching cost computation in stereo matching. However, conventional CNN-based approaches to learn the network in a supervised manner require a large number of ground-truth disparity maps, which limits their applicability. To overcome this limitation, we present a novel framework to learn a CNNs architecture for matching cost computation in an unsupervised manner. Our method leverages an image domain learning combined with stereo epipolar constraints. Exploiting the correspondence consistency between stereo images as supervision, our method selects the training samples in each iteration during network training and uses them to learn the network. To boost the performance, we also propose a multi-scale cost computation scheme. Experimental results show that our method outperforms the state-of-the-art methods including even supervised learning based methods on various benchmarks.

AB - Deep convolutional neural networks (CNNs) have shown revolutionary performance improvements for matching cost computation in stereo matching. However, conventional CNN-based approaches to learn the network in a supervised manner require a large number of ground-truth disparity maps, which limits their applicability. To overcome this limitation, we present a novel framework to learn a CNNs architecture for matching cost computation in an unsupervised manner. Our method leverages an image domain learning combined with stereo epipolar constraints. Exploiting the correspondence consistency between stereo images as supervision, our method selects the training samples in each iteration during network training and uses them to learn the network. To boost the performance, we also propose a multi-scale cost computation scheme. Experimental results show that our method outperforms the state-of-the-art methods including even supervised learning based methods on various benchmarks.

UR - http://www.scopus.com/inward/record.url?scp=85045303743&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85045303743&partnerID=8YFLogxK

U2 - 10.1109/ICIP.2017.8296736

DO - 10.1109/ICIP.2017.8296736

M3 - Conference contribution

T3 - Proceedings - International Conference on Image Processing, ICIP

SP - 2518

EP - 2522

BT - 2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings

PB - IEEE Computer Society

ER -

Joung S, Kim S, Ham B, Sohn K. Unsupervised stereo matching using correspondence consistency. In 2017 IEEE International Conference on Image Processing, ICIP 2017 - Proceedings. IEEE Computer Society. 2018. p. 2518-2522. (Proceedings - International Conference on Image Processing, ICIP). https://doi.org/10.1109/ICIP.2017.8296736