What Makes a Good Detector? – Structured Priors for Learning from Few Examples

  • Tianshi Gao
  • Michael Stark
  • Daphne Koller
Part of the Lecture Notes in Computer Science book series (LNCS, volume 7576)


Transfer learning can counter the heavy-tailed nature of the distribution of training examples over object classes. Here, we study transfer learning for object class detection. Starting from the intuition that “what makes a good detector” should manifest itself in the form of repeatable statistics over existing “good” detectors, we design a low-level feature model that can be used as a prior for learning new object class models from scarce training data. Our priors are structured, capturing dependencies both on the level of individual features and spatially neighboring pairs of features. We confirm experimentally the connection between the information captured by our priors and “good” detectors as well as the connection to transfer learning from sources of different quality. We give an in-depth analysis of our priors on a subset of the challenging PASCAL VOC 2007 data set and demonstrate improved average performance over all 20 classes, achieved without manual intervention.


Training Image Semantic Relatedness Object Class Target Class Transfer Learning 
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.
    Everingham, M., van Gool, L., Williams, C.K., Winn, J., Zisserman, A.: The pascal visual object classes (voc) challenge. In: IJCV (2010)Google Scholar
  2. 2.
    Wang, G., Forsyth, D., Hoiem, D.: Comparative object similarity for improved recognition with few or no examples. In: CVPR (2010)Google Scholar
  3. 3.
    Salakhutdinov, R., Torralba, A., Tenenbaum, J.: Learning to share visual appearance for multiclass object detection. In: CVPR (2011)Google Scholar
  4. 4.
    Lim, J.J., Salakhutdinov, R., Torralba, A.: Transfer learning by borrowing examples for multiclass object detection. In: NIPS (2011)Google Scholar
  5. 5.
    Torralba, A., Murphy, K., Freeman, W.: Sharing visual features for multiclass and multiview object detection. In: CVPR (2004)Google Scholar
  6. 6.
    Luo, J., Tommasi, T., Caputo, B.: Multiclass transfer learning from unconstrained priors. In: ICCV (2011)Google Scholar
  7. 7.
    Lampert, C.H., Nickisch, H., Harmeling, S.: Learning to detect unseen object classes by between-class attribute transfer. In: CVPR (2009)Google Scholar
  8. 8.
    Farhadi, A., Endres, I., Hoiem, D.: Attribute-centric recognition for cross-category generalization. In: CVPR (2010)Google Scholar
  9. 9.
    Wang, Y., Mori, G.: A Discriminative Latent Model of Object Classes and Attributes. In: Daniilidis, K., Maragos, P., Paragios, N. (eds.) ECCV 2010, Part V. LNCS, vol. 6315, pp. 155–168. Springer, Heidelberg (2010)CrossRefGoogle Scholar
  10. 10.
    Rohrbach, M., Stark, M., Schiele, B.: Evaluating knowledge transfer and zero-shot learning in a large-scale setting. In: CVPR (2011)Google Scholar
  11. 11.
    Levi, K., Fink, M., Weiss, Y.: Learning from a small number of training examples by exploiting object categories. In: LCVPR (2004)Google Scholar
  12. 12.
    Zweig, A., Weinshall, D.: Exploiting object hierarchy: Combining models from different category levels. In: ICCV (2007)Google Scholar
  13. 13.
    Li, L.J., Wang, C., Lim, Y., Blei, D., Fei-Fei, L.: Building and using a semantivisual image hierarchy. In: CVPR (2010)Google Scholar
  14. 14.
    Dalal, N., Triggs, B.: Histograms of oriented gradients for human detection. In: CVPR (2005)Google Scholar
  15. 15.
    Felzenszwalb, P., Girshick, R., McAllester, D., Ramanan, D.: Object detection with discriminatively trained part based models. PAMI (2010)Google Scholar
  16. 16.
    Aytar, Y., Zisserman, A.: Tabula rasa: Model transfer for object category detection. In: ICCV (2011)Google Scholar
  17. 17.
    Stark, M., Goesele, M., Schiele, B.: A shape-based object class model for knowledge transfer. In: ICCV (2009)Google Scholar
  18. 18.
    Bart, E., Ullman, S.: Cross-generalization: Learning novel classes from a single example by feature replacement. In: CVPR (2005)Google Scholar
  19. 19.
    Ferrari, V., Zisserman, A.: Learning visual attributes. In: NIPS (2007)Google Scholar
  20. 20.
    Berg, T.L., Berg, A.C., Shih, J.: Automatic Attribute Discovery and Characterization from Noisy Web Data. In: Daniilidis, K., Maragos, P., Paragios, N. (eds.) ECCV 2010, Part I. LNCS, vol. 6311, pp. 663–676. Springer, Heidelberg (2010)CrossRefGoogle Scholar
  21. 21.
    Tommasi, T., Orabona, F., Caputo, B.: Safety in numbers: Learning categories from few examples with multi model knowledge transfer. In: CVPR (2010)Google Scholar
  22. 22.
    Marszalek, M., Schmid, C.: Semantic hierarchies for visual object recognition. In: CVPR (2007)Google Scholar
  23. 23.
    Bart, E., Ullman, S.: Single-example learning of novel classes using representation by similarity. In: BMVC (2005)Google Scholar
  24. 24.
    Miller, E., Matsakis, N., Viola, P.: Learning from One Example Through Shared Densities on Transforms. In: CVPR (2000)Google Scholar
  25. 25.
    Fei-Fei, L., Fergus, R., Perona, P.: One-shot learning of object categories. PAMI 28, 594–611 (2006)CrossRefGoogle Scholar
  26. 26.
    Raina, R., Ng, A.Y., Koller, D.: Constructing informative priors using transfer learning. In: ICML (2006)Google Scholar
  27. 27.
    Elidan, G., Packer, B., Heitz, G., Koller, D.: Convex point estimation using undirected bayesian transfer hierarchies. In: UAI (2008)Google Scholar
  28. 28.
    Koller, D., Friedman, N.: Probabilistic Graphical Models: Principles and Techniques. MIT Press (2009)Google Scholar
  29. 29.
    Fellbaum, C.: WordNet: An Electronical Lexical Database. The MIT Press (1998)Google Scholar
  30. 30.
    Gabrilovich, E., Markovitch, S.: Computing Semantic Relatedness using Wikipedia-based Explicit Semantic Analysis. In: IJCAI (2007)Google Scholar

Copyright information

© Springer-Verlag Berlin Heidelberg 2012

Authors and Affiliations

  • Tianshi Gao
    • 1
  • Michael Stark
    • 1
    • 2
  • Daphne Koller
    • 1
  1. 1.Stanford UniversityUSA
  2. 2.Max Planck Institute for InformaticsGermany

Personalised recommendations