CSc 665

Advanced Topics in Computational Intelligence (fall 2010):
Computer vision research projects (Draft!)

Link to Class Schedule

Link to Project Suggestions

Supplementary material

Instructor Kobus Barnard
Office Hours TBA, by automated sign up.
Time and Place Friday 9:30-12:00, Gould-Simpson 942
Description This course will be an intense research focused seminar course that will combine computer vision content with significant project work, paper writing, and presentation. The computer vision content will be chosen to compliment what is covered in CS 577. (Students who have not taken CS 577 should see "prerequisites" below). Presentation of the content will be handled largely by students. The instructor will provide short lectures on research methodology and effective writing and presentation.

Some particulars will have to wait until the first week of classes as they depend on the class size and student interest.

Technical Content We will develop a collection of topics and classic papers as the class progresses, using an initial set provided by the instructor as a starting point. Students will present the material in a tutorial fashion for roughly half of the overall class time. While all students will study the material, and topic presenters will assign reading as appropriate, the presenters will be challenged to teach the material. Lectures will followed by short group feedback sessions. Example topics (under construction) include:
  • Probabilistic tracking
  • Multiple view geometry and stereo
  • Shape representation and generative structure models
  • Recognition by appearance matching
  • Correspondance for recognition from pose consistancy
  • Image and video retrieval
  • Vision and language (AKA words and pictures)
  • Project Work Much of the course will revolve around student research projects. Research projects must have some connection with computer vison, but can have substantive aspects from related fields such as machine learning, graphics, multimedia, robotics, or cognitive science. We will assume sufficient prior exposure to computer vision or related fields so that projects can be started very early in the term. Project selection will be finalized within the first three weeks. A list of project ideas (under construction) can be found here.   Projects can emerge from that list, students own research, vision lab research, or needed computer vision work around campus. Projects must contribute to a research effort or infrastructure, either new or ongoing, within the vision lab or external to it. Projects that have no use after the class ends are not suitable for this course.

    Many possible projects are "encumbered" in that they are in collaboration with ongoing efforts, and will potentially require sharing of authorship and software products, as well as requirements on how software is developed (e.g., language and platforms). Students will have to take these issues into account when choosing a project.

    Group projects are possible, but each student needs to partition off a conceptual piece of the work, and write their own paper. Sharing of material such as introductory text is not permissible --- papers from collaborating projects should cite each other. How the collaboration will work needs to be addressed in the research proposal. Similarly, projects that continue or extend ongoing research is possible, but how this is a separate piece of work may need to be addressed in the proposal.

    Sharing of infrastructure (e.g., software, ground truth data) is strongly encouraged, both among collaborating students and in the group as a whole. In fact, contributions to usable research infrastructure will contribute to the project impact marks and/or bonus marks, as well as research karma.

    Multiple aspects of the project will undergo "peer review", patterned on grant and research paper reviewing. In addition, project progress will be presented in class. For a tentative schedule of due dates, follow this link.

    Prerequisites A previous computer vision course (e.g., CS 577), or research experience in computer vision or a related areas (e.g., machine learning or graphics), or permission from the instructor. We will hit the ground running doing computer vision, and thus it is critical that students have both the motivation and background to do so. In particular, familiarity with the following concepts will be assumed and thus need to either be reviewed or learned in a hurry:
  • Projective cameras, modeled by linear transormation in homogeneous coordinates, and what it means to calibrate such a camera (e.g., F&P chapter 2.2).
  • Extraction of low level localized features representing color and texture
  • Extraction of scale and rotations invariant feature descriptors at highly informative image areas (e.g., Lowe's 2004 paper).
  • Basic clustering (e.g, k-means) and binary classification (e.g., support vector machines).
  • Required Text There are no required texts. All readings will be made available. However, much material will follow "Computer Vision: A Modern Approach," by Forsyth and Ponce, which is the recommended general reference this course. The computer vision lab has a few copies of this book as well as other relevant volumes that can be signed out for a few days.

    Grading will be based on performance in the following areas (details subject to change).

    • Participation (10%)
    • Lectures (20%)
    • Proposal (10%)
      • Proposal presentation (3%)
      • Proposal draft (5%)
      • Final proposal (2%)
    • Project (45%)
      • Project progress presentation (5%)
      • Paper draft (15%)
      • Final project presentation (5%)
      • Project impact (15%)
      • Final paper (5%)
    • Reviews (15%)
      • Proposal draft review (3%)
      • Paper draft review (7%)
      • Final paper review (5%)

    A cumulative percentage of 90% guarantees an A, 80% guarantees a B, 70% a C, and 60% a D.

    Policies Good attendance is required. If you cannot make class due to travel or sickness, please let the instructor know, as missed classes can count against the participation grade. The degree of impact will be significant if there are more than two unexplained missed classes.

    Students will be expected to respect the University of Arizona's academic integrity policy.

    Class Schedule