On learning to localize objects with minimal supervision

Hyun Oh Song, Ross Girshick, Stefanie Jegelka, Julien Mairal, Zaid Harchaoui, Trevor Darrell
Proceedings of the 31st International Conference on Machine Learning, PMLR 32(2):1611-1619, 2014.

Abstract

Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50% relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection.

Cite this Paper


BibTeX
@InProceedings{pmlr-v32-songb14, title = {On learning to localize objects with minimal supervision}, author = {Song, Hyun Oh and Girshick, Ross and Jegelka, Stefanie and Mairal, Julien and Harchaoui, Zaid and Darrell, Trevor}, booktitle = {Proceedings of the 31st International Conference on Machine Learning}, pages = {1611--1619}, year = {2014}, editor = {Xing, Eric P. and Jebara, Tony}, volume = {32}, number = {2}, series = {Proceedings of Machine Learning Research}, address = {Bejing, China}, month = {22--24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v32/songb14.pdf}, url = {https://proceedings.mlr.press/v32/songb14.html}, abstract = {Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50% relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection.} }
Endnote
%0 Conference Paper %T On learning to localize objects with minimal supervision %A Hyun Oh Song %A Ross Girshick %A Stefanie Jegelka %A Julien Mairal %A Zaid Harchaoui %A Trevor Darrell %B Proceedings of the 31st International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2014 %E Eric P. Xing %E Tony Jebara %F pmlr-v32-songb14 %I PMLR %P 1611--1619 %U https://proceedings.mlr.press/v32/songb14.html %V 32 %N 2 %X Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50% relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection.
RIS
TY - CPAPER TI - On learning to localize objects with minimal supervision AU - Hyun Oh Song AU - Ross Girshick AU - Stefanie Jegelka AU - Julien Mairal AU - Zaid Harchaoui AU - Trevor Darrell BT - Proceedings of the 31st International Conference on Machine Learning DA - 2014/06/18 ED - Eric P. Xing ED - Tony Jebara ID - pmlr-v32-songb14 PB - PMLR DP - Proceedings of Machine Learning Research VL - 32 IS - 2 SP - 1611 EP - 1619 L1 - http://proceedings.mlr.press/v32/songb14.pdf UR - https://proceedings.mlr.press/v32/songb14.html AB - Learning to localize objects with minimal supervision is an important problem in computer vision, since large fully annotated datasets are extremely costly to obtain. In this paper, we propose a new method that achieves this goal with only image-level labels of whether the objects are present or not. Our approach combines a discriminative submodular cover problem for automatically discovering a set of positive object windows with a smoothed latent SVM formulation. The latter allows us to leverage efficient quasi-Newton optimization techniques. Our experiments demonstrate that the proposed approach provides a 50% relative improvement in mean average precision over the current state-of-the-art on PASCAL VOC 2007 detection. ER -
APA
Song, H.O., Girshick, R., Jegelka, S., Mairal, J., Harchaoui, Z. & Darrell, T.. (2014). On learning to localize objects with minimal supervision. Proceedings of the 31st International Conference on Machine Learning, in Proceedings of Machine Learning Research 32(2):1611-1619 Available from https://proceedings.mlr.press/v32/songb14.html.

Related Material