Weakly Supervised Object Localization with Progressive Domain Adaptation

Dong Li, Jia Bin Huang, Yali Li, Shengjin Wang, Ming Hsuan Yang

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

54 Citations (Scopus)

Abstract

We address the problem of weakly supervised object localization where only image-level annotations are available for training. Many existing approaches tackle this problem through object proposal mining. However, a substantial amount of noise in object proposals causes ambiguities for learning discriminative object models. Such approaches are sensitive to model initialization and often converge to an undesirable local minimum. In this paper, we address this problem by progressive domain adaptation with two main steps: classification adaptation and detection adaptation. In classification adaptation, we transfer a pre-trained network to our multi-label classification task for recognizing the presence of a certain object in an image. In detection adaptation, we first use a mask-out strategy to collect class-specific object proposals and apply multiple instance learning to mine confident candidates. We then use these selected object proposals to fine-tune all the layers, resulting in a fully adapted detection network. We extensively evaluate the localization performance on the PASCAL VOC and ILSVRC datasets and demonstrate significant performance improvement over the state-of-the-art methods.

Original languageEnglish
Title of host publicationProceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016
PublisherIEEE Computer Society
Pages3512-3520
Number of pages9
ISBN (Electronic)9781467388504
DOIs
Publication statusPublished - 2016 Dec 9
Event29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016 - Las Vegas, United States
Duration: 2016 Jun 262016 Jul 1

Publication series

NameProceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition
Volume2016-December
ISSN (Print)1063-6919

Conference

Conference29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016
CountryUnited States
CityLas Vegas
Period16/6/2616/7/1

Fingerprint

Volatile organic compounds
Labels
Masks

All Science Journal Classification (ASJC) codes

  • Software
  • Computer Vision and Pattern Recognition

Cite this

Li, D., Huang, J. B., Li, Y., Wang, S., & Yang, M. H. (2016). Weakly Supervised Object Localization with Progressive Domain Adaptation. In Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016 (pp. 3512-3520). [7780751] (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition; Vol. 2016-December). IEEE Computer Society. https://doi.org/10.1109/CVPR.2016.382
Li, Dong ; Huang, Jia Bin ; Li, Yali ; Wang, Shengjin ; Yang, Ming Hsuan. / Weakly Supervised Object Localization with Progressive Domain Adaptation. Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016. IEEE Computer Society, 2016. pp. 3512-3520 (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition).
@inproceedings{0197e099b9514c2eafe538cd38b6e09d,
title = "Weakly Supervised Object Localization with Progressive Domain Adaptation",
abstract = "We address the problem of weakly supervised object localization where only image-level annotations are available for training. Many existing approaches tackle this problem through object proposal mining. However, a substantial amount of noise in object proposals causes ambiguities for learning discriminative object models. Such approaches are sensitive to model initialization and often converge to an undesirable local minimum. In this paper, we address this problem by progressive domain adaptation with two main steps: classification adaptation and detection adaptation. In classification adaptation, we transfer a pre-trained network to our multi-label classification task for recognizing the presence of a certain object in an image. In detection adaptation, we first use a mask-out strategy to collect class-specific object proposals and apply multiple instance learning to mine confident candidates. We then use these selected object proposals to fine-tune all the layers, resulting in a fully adapted detection network. We extensively evaluate the localization performance on the PASCAL VOC and ILSVRC datasets and demonstrate significant performance improvement over the state-of-the-art methods.",
author = "Dong Li and Huang, {Jia Bin} and Yali Li and Shengjin Wang and Yang, {Ming Hsuan}",
year = "2016",
month = "12",
day = "9",
doi = "10.1109/CVPR.2016.382",
language = "English",
series = "Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition",
publisher = "IEEE Computer Society",
pages = "3512--3520",
booktitle = "Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016",
address = "United States",

}

Li, D, Huang, JB, Li, Y, Wang, S & Yang, MH 2016, Weakly Supervised Object Localization with Progressive Domain Adaptation. in Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016., 7780751, Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, vol. 2016-December, IEEE Computer Society, pp. 3512-3520, 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016, Las Vegas, United States, 16/6/26. https://doi.org/10.1109/CVPR.2016.382

Weakly Supervised Object Localization with Progressive Domain Adaptation. / Li, Dong; Huang, Jia Bin; Li, Yali; Wang, Shengjin; Yang, Ming Hsuan.

Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016. IEEE Computer Society, 2016. p. 3512-3520 7780751 (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition; Vol. 2016-December).

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

TY - GEN

T1 - Weakly Supervised Object Localization with Progressive Domain Adaptation

AU - Li, Dong

AU - Huang, Jia Bin

AU - Li, Yali

AU - Wang, Shengjin

AU - Yang, Ming Hsuan

PY - 2016/12/9

Y1 - 2016/12/9

N2 - We address the problem of weakly supervised object localization where only image-level annotations are available for training. Many existing approaches tackle this problem through object proposal mining. However, a substantial amount of noise in object proposals causes ambiguities for learning discriminative object models. Such approaches are sensitive to model initialization and often converge to an undesirable local minimum. In this paper, we address this problem by progressive domain adaptation with two main steps: classification adaptation and detection adaptation. In classification adaptation, we transfer a pre-trained network to our multi-label classification task for recognizing the presence of a certain object in an image. In detection adaptation, we first use a mask-out strategy to collect class-specific object proposals and apply multiple instance learning to mine confident candidates. We then use these selected object proposals to fine-tune all the layers, resulting in a fully adapted detection network. We extensively evaluate the localization performance on the PASCAL VOC and ILSVRC datasets and demonstrate significant performance improvement over the state-of-the-art methods.

AB - We address the problem of weakly supervised object localization where only image-level annotations are available for training. Many existing approaches tackle this problem through object proposal mining. However, a substantial amount of noise in object proposals causes ambiguities for learning discriminative object models. Such approaches are sensitive to model initialization and often converge to an undesirable local minimum. In this paper, we address this problem by progressive domain adaptation with two main steps: classification adaptation and detection adaptation. In classification adaptation, we transfer a pre-trained network to our multi-label classification task for recognizing the presence of a certain object in an image. In detection adaptation, we first use a mask-out strategy to collect class-specific object proposals and apply multiple instance learning to mine confident candidates. We then use these selected object proposals to fine-tune all the layers, resulting in a fully adapted detection network. We extensively evaluate the localization performance on the PASCAL VOC and ILSVRC datasets and demonstrate significant performance improvement over the state-of-the-art methods.

UR - http://www.scopus.com/inward/record.url?scp=84986317248&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84986317248&partnerID=8YFLogxK

U2 - 10.1109/CVPR.2016.382

DO - 10.1109/CVPR.2016.382

M3 - Conference contribution

AN - SCOPUS:84986317248

T3 - Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition

SP - 3512

EP - 3520

BT - Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016

PB - IEEE Computer Society

ER -

Li D, Huang JB, Li Y, Wang S, Yang MH. Weakly Supervised Object Localization with Progressive Domain Adaptation. In Proceedings - 29th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2016. IEEE Computer Society. 2016. p. 3512-3520. 7780751. (Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition). https://doi.org/10.1109/CVPR.2016.382