Non Maximal Suppression in Cascaded Ranking Models

  • Matthew B. Blaschko
  • Juho Kannala
  • Esa Rahtu
Part of the Lecture Notes in Computer Science book series (LNCS, volume 7944)


Ranking models have recently been proposed for cascaded object detection, and have been shown to improve over regression or binary classification in this setting [1,2]. Rather than train a classifier in a binary setting and interpret the function post hoc as a ranking objective, these approaches directly optimize regularized risk objectives that seek to score highest the windows that most closely match the ground truth. In this work, we evaluate the effect of non-maximal suppression (NMS) on the cascade architecture, showing that this step is essential for high performance. Furthermore, we demonstrate that non-maximal suppression has a significant effect on the tradeoff between recall different points on the overlap-recall curve. We further develop additional objectness features at low computational cost that improve performance on the category independent object detection task introduced by Alexe et al. [3]. We show empirically on the PASCAL VOC dataset that a simple and efficient NMS strategy yields better results in a typical cascaded detection architecture than the previous state of the art [4.1]. This demonstrates that NMS, an often ignored stage in the detection pipeline, can be a dominating factor in the performance of detection systems.


Object Detection Image Patch Preference Graph Detection Architecture Ranking Objective 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


  1. 1.
    Rahtu, E., Kannala, J., Blaschko, M.B.: Learning a category independent object detection cascade. In: Proc. ICCV (2011)Google Scholar
  2. 2.
    Zhang, Z., Warrell, J., Torr, P.H.S.: Proposal generation for object detection using cascaded ranking SVMs. In: Proc. CVPR (2011)Google Scholar
  3. 3.
    Alexe, B., Deselaers, T., Ferrari, V.: What is an object. In: Proc. CVPR (2010)Google Scholar
  4. 4.
    Alexe, B., Deselaers, T., Ferrari, V.: Measuring the objectness of image windows. IEEE PAMI (2012)Google Scholar
  5. 5.
    Viola, P., Jones, M.: Robust real-time object detection. IJCV 1 (2001)Google Scholar
  6. 6.
    Romdhani, S., Torr, P., Schölkopf, B., Blake, A.: Computationally efficient face detection. In: Proc. ICCV (2001)Google Scholar
  7. 7.
    Vedaldi, A., Gulshan, V., Varma, M., Zisserman, A.: Multiple kernels for object detection. In: Proc. ICCV (2009)Google Scholar
  8. 8.
    Wu, J., Brubaker, S.C., Mullin, M.D., Rehg, J.M.: Fast asymmetric learning for cascade face detection. IEEE PAMI 30, 369–382 (2008)CrossRefGoogle Scholar
  9. 9.
    Lampert, C.H., Blaschko, M.B., Hofmann, T.: Efficient subwindow search: A branch and bound framework for object localization. IEEE PAMI (2009)Google Scholar
  10. 10.
    Lehmann, A., Gehler, P., Van Gool, L.: Branch & rank: Non-linear object detection. In: Proc. BMVC (2011)Google Scholar
  11. 11.
    Herbrich, R., Graepel, T., Obermayer, K.: Large margin rank boundaries for ordinal regression. In: Smola, A.J., Bartlett, P.L., Schölkopf, B., Schuurmans, D. (eds.) Advances in Large Margin Classifiers, pp. 115–132. MIT Press (2000)Google Scholar
  12. 12.
    Bakir, G.H., Hofmann, T., Schölkopf, B., Smola, A.J., Taskar, B., Vishwanathan, S.V.N.: Predicting structured data. MIT Press (2007)Google Scholar
  13. 13.
    Blaschko, M.B., Vedaldi, A., Zisserman, A.: Simultaneous object detection and ranking with weak supervision. In: NIPS (2010)Google Scholar
  14. 14.
    Mittal, A., Blaschko, M.B., Zisserman, A., Torr, P.H.S.: Taxonomic multi-class prediction and person layout using efficient structured ranking. In: Fitzgibbon, A., Lazebnik, S., Perona, P., Sato, Y., Schmid, C. (eds.) ECCV 2012, Part II. LNCS, vol. 7573, pp. 245–258. Springer, Heidelberg (2012)CrossRefGoogle Scholar
  15. 15.
    Barinova, O., Lempitsky, V., Kohli, P.: On the detection of multiple object instances using Hough transforms. In: Proc. CVPR (2010)Google Scholar
  16. 16.
    Blaschko, M.B.: Branch and bound strategies for non-maximal suppression in object detection. In: Boykov, Y., Kahl, F., Lempitsky, V., Schmidt, F.R. (eds.) EMMCVPR 2011. LNCS, vol. 6819, pp. 385–398. Springer, Heidelberg (2011)Google Scholar
  17. 17.
    McAllester, D.: Generalization bounds and consistency for structured labeling. In: Bakır, G.H., Hofmann, T., Schölkopf, B., Smola, A.J., Taskar, B., Vishwanathan, S.V.N. (eds.) Predicting Structured Data, pp. 247–261. MIT Press (2007)Google Scholar
  18. 18.
    Nemhauser, G.L., Wolsey, L.A., Fisher, M.L.: An analysis of approximations for maximizing submodular set functions. Mathematical Programming 14, 265–294 (1978)MathSciNetzbMATHCrossRefGoogle Scholar
  19. 19.
    Everingham, M., Van Gool, L., Williams, C.K.I., Winn, J., Zisserman, A.: The PASCAL Visual Object Classes (VOC) challenge. IJCV 88(2), 303–338 (2010)CrossRefGoogle Scholar
  20. 20.
    Hyvärinen, A., Hurri, J., Hoyer, P.: Natural Image Statistics. Springer (2009)Google Scholar
  21. 21.
    Deselaers, T., Alexe, B., Ferrari, V.: Localizing objects while learning their appearance. In: Daniilidis, K., Maragos, P., Paragios, N. (eds.) ECCV 2010, Part IV. LNCS, vol. 6314, pp. 452–466. Springer, Heidelberg (2010)CrossRefGoogle Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2013

Authors and Affiliations

  • Matthew B. Blaschko
    • 1
    • 2
    • 3
  • Juho Kannala
    • 4
  • Esa Rahtu
    • 4
  1. 1.Center for Visual ComputingÉcole Centrale ParisFrance
  2. 2.Équipe Galen, INRIA Saclay, Île-de-FranceFrance
  3. 3.LIGM (UMR CNRS), École des Ponts ParisTechUniversité Paris-EstFrance
  4. 4.Machine Vision GroupUniversity of OuluFinland

Personalised recommendations