Feature Descriptors for Shapes and Images
WS 2011/12, TU München
Location: Room 02.09.023
Date: March 5 and 6
Time: To be determined
Lecturer: Mathieu Aubry
The course will be held in English. The student can do his / her presentation in German, but English is prefered.
In this seminar students will present some important papers about local descriptors for images and shapes. We will focus mainly on three topics:
- descriptors for images (SIFT, SURF, Hog…),
- descriptor learning,
- descriptors for shapes (HKS…).
The student will have some freedom in the choice of the content and time of their presentations. The only constraints are that they must present the main idea(s) of the paper in a way that all students attending the course can understand it and that the presentation should be approximately 45 minutes long. The structure of your work and the way you present the content is up to you, but must be structured and logical. You can develop the theoretical content of the paper, or present some experiments/test of your own (the code is freely available for many papers), or do some comparisons with other methods. It is strongly advised to have at least a ppt support for part of the presentation, but you can choose to do part of it at the blackboard and using hand-outs for the participants.
I will meet each student, at least one time, at least two weeks before the presentation. Two weeks before the presentation, the paper should be well understood and a draft version of the presentation available. In the meeting, we will go over the paper and review the presentation together. Note, that you can contact me anytime before and ask me questions at any point.
The presentations will be followed by questions and discussions in class. The evaluation will be based on the quality of the presentation (and take into account the difficulty of the paper and the initial level of the student) and on the participation in the discussions.
Attendance is limited to 15 students (Bachelor or Master).
For more information or to sign up please contact me at email@example.com.
All the initial papers have been taken. You can still join the seminar: in this case just write me an e-mail precising your area of interest and I will suggest more papers.
descriptors for images
Distinctive image features from scale-invariant keypoints, Lowe, D.G., International journal of computer vision 2004.
Surf: Speeded up robust features, Bay, H. and Tuytelaars, T. and Van Gool, L., ECCV 2006.
Histograms of oriented gradients for human detection,Dalal, N. and Triggs, B., CVPR 2005
Brief: Binary robust independent elementary features, Calonder, M. and Lepetit, V. and Strecha, C. and Fua, P., ECCV 2010
descriptors for shapes
A Concise and Provably Informative Multi-Scale Signature Based on Heat Diffusion, Sun, J. and Ovsjanikov, M. and Guibas, L., Computer Graphics Forum 2009
Learning mid-level features for recognition, Boureau, Y.L. and Bach, F. and LeCun, Y. and Ponce, J., CVPR 2010
Keypoint signatures for fast learning and recognition, Calonder, M. and Lepetit, V. and Fua, P., ECCV 2008
If you have a specific interest concerning some other paper or technique, the scope of the seminar can be extended to the corresponding area and topic, e.g. more towards keypoint detection, the use of descriptors for retrieval or matching, or others.