Göm meny

Visual Object Recognition HT2014 (6hp+2hp)

This course introduces the participants to state of the art in visual object recognition, with a focus on recent algorithms with good practical utility. The course consists of eight lectures, a collection of scientific papers to read, and a written exam. The main course is worth 6hp. Extra credits (2hp) are awarded if a small project is also implemented.

  • Each lecture (except the first) is accompanied by a scientific paper that should be read in advance.
  • For the optional small project, you should preferrably do a project related to your own research. A project should consist of implementing a particular algorithm, or doing a proper evaluation of an existing implementation. Before starting a project you should get it approved.

Learning Goals

After the course, you should have a good grasp of the research front in visual object recognition. You should also be able to explain algorithms for the following topics: invariant frames, feature detectors, descriptor construction, matching, tree search, and performance evaluation.

The Lectures

Course start is December 9, 13-15, in Signalen.

The seminars will be spaced one week apart, and most of them are in 2015. Please bring your calendar for the first seminar, in case we need to move some seminars (only if there is a collision for many participants).

The lectures/seminars are 120min each, and will be held at the Department of Electrical Engineering.

  1. Introduction [PDF]
    Dec 9, 13-15, in Signalen.
    Terminology: recognition, classification, categorisation, detection, pose estimation, expression. Learning for recognition. Local vs. holistic features. Application examples.
  2. Image formation [PDF]
    Jan 27, 12.30-15, in Signalen.
    Pin-hole and thin lens models. Epipolar geometry. Feature invariances, illumination, colour, homographies, canonical frames.
  3. Descriptors [PDF]
    Feb 3, 12.30-15, in Signalen.
    HOG/SIFT, BRIEF, Ferns, Texture descriptors, Shape descriptors, Colour histograms, GIST, descriptor learning.
  4. Region Detectors [PDF]
    Feb 10, 12.30-15, in Signalen.
    DoG, Harris, FAST, MSER, MSCR, scale selection, affine adaptation.
  5. Compound descriptors and metrics [PDF]
    Feb 24, 12.30-15, in Signalen.
    visual words, bags-of-features, spatial pyramids, Constellations, Ratio score, Chi2 distance, metric learning.
  6. Tree search and hashing algorithms [PDF]
    March 10, 12.30-15, in Signalen.
    High dimensional spaces, kD-trees, BBF, ball trees, k-means tree, FLANN. Geometric hashing, Locality Sensitive Hashing.
  7. Voting and learning [PDF]
    March 25, 12.30-15, in Systemet.
    Generalized hough transform. Meanshift Voting, Random Forests, Deep Learning, Convolutional Networks.
  8. Performance evaluation [PDF]
    April 7, 12.30-15, in Signalen.
    ImageNet LSVRC. Repeatability tests, inlier-frequency curve, precision-recall and ROC curves. etc.

Seminar Articles

Articles to read before seminars 2-8 are listed below.

  1. M. Brown, D.G. Lowe, "Invariant Features from Interest Point Groups", BMVC02 [link to PDF]
  2. M. Calonder et al., "BRIEF: Binary Robust Independent Elementary Features", ECCV10 [link to PDF]
  3. S. Leutenegger et al., "BRISK: Binary Robust Invariant Scalable Keypoints", ICCV11 [link to PDF]
  4. J. Sivic, A. Zisserman, "Video Google: A Text Retrieval Approach to Object Matching in Videos", ICCV03 [link to PDF]
  5. M. Muja, D.G. Lowe, "Scalable Nearest Neighbour Algorithms for High Dimensional Data", TPAMI14 [link to PDF]
  6. A. Krizhevsky et al., "ImageNet Classification with Deep Convolutional Neural Networks", NIPS12 [link to PDF]
  7. Russakovsky and Deng et al., "ImageNet Large Scale Visual Recognition Challenge", ArXiV15 [link to PDF]

Other recommended articles:

  1. K. Mikolajczyk, et al., "A Comparison of Affine Region Detectors", Springer IJCV 2006
  2. D. G. Lowe, "Distinctive Features from Scale-Invariant Keypoints", Springer IJCV 2004
  3. J. Davis and M. Goadrich, "The Relationship Between Precision-Recall and ROC Curves", ICML 2006

Registration

  • If you are interested in participating, you should register by sending an email to Per-Erik Forssén before December 1. email: perfo AT isy.liu.se.
  • The email should contain: (1) full name, (2) personal number, (3) department or equivalent, (4) name of your supervisor.
  • Note that even if you just want to attend the seminars you should still register, so I can book a room of the right size.

Senast uppdaterad: 2015-04-07