CPS 296.1
Introduction to Computer Vision
Spring 2003

Project Guidelines

Students taking this course for credit will work on a computer vision project. Projects can be done independently or in groups, but group projects are strongly encouraged. Some suggestions about topics are given below. There is no specific format for a project, but some guidelines are given below to ensure a certain degree of uniformity in the effort required. These guidelines include checkpoints when you have to hand in material. Make sure you read this document in its entirety.

Examples

It is best if you come up with your own idea. However, here are some project types:

• Project ideas are given in the form of programming assignments at the end of every chapter in the textbook. These projects are likely to be successful, because you will know the relevant material.
• Web-based exploration of a vision topic. Search the web, perhaps starting from CVonline, for pages on camera calibration, stereo vision, edge detection, tracking people, 3D reconstruction, etc. Find a couple of algorithms that come with code and sample input data. Run the algorithms on all inputs, and find ways to compare their performance. Success here is tied to a well-designed comparison procedure, and a clear analysis and presentation of the results.
• Modify an existing algorithm. Choose a program from the Web (for instance, an edge detector). Run the program on some images, and determine some shortcoming of it. Propose a modification for improving the algorithm, which you then implement and test. This project is successful if you complete your implementation and discuss the testing results. It does not matter if these results are not as good as expected. It is even OK if your modification makes the results worse, as long as you show this and make an honest effort to understand why.
• Write your own implementation of a vision algorithm that already exists in the literature. This could be about stereo matching, tracking, calibration, edge detection, or any other topic of interest. A place to start is to look at something that attracts you in the textbook, read the relevant page or two, and pick one of the references. A project of this type is successful if your code works and you can run experiments and discuss strengths and drawbacks.
• Present your own solution to a vision problem: Can you think of a novel solution to the correspondence problem? Is there a noise-insensitive reconstruction from image sequences? This is a very high risk type of project, definitely not recommended unless you already have a concrete idea in mind and/or you have played with it in the past.

 

As you see, programming is not an absolute requirement. This is a vision class, not a programming class.

Checkpoints

Some checkpoints are established throughout the quarter to make sure that you get early feedback on your work. The checkpoints are as follows.

 

February 4, 2003

Project proposal is due (on-line, before class)
(20 percent of project grade)

March 25, 2003

Project review is due (on-line, before class)
(30 percent of project grade)

April 10, 2003

Project's final report is due (on-line, before class)
(50 percent of project grade)

 

All report materials should be in HTML format (more on this below). Either e-mail a URL to tomasi@cs.duke.edu that points to your HTML files, or else send me the HTML files themselves in an e-mail. If you use the latter method, please send me a single zip archive file. Each of the checkpoints involve handing in material on your part. This material will be graded, and all deadlines are strict. Only under special circumstances, discussed with me in advance, can the deadline be moved by a few days. These checkpoints are discussed in more detail in the remainder of this document.

Project Proposal

The project proposal is a one page description of what you intend to do. It should be an HTML document. The reason for this format is that your projects will be published on the World-Wide Web. WWW pages can be read from across the world, so please do a careful job writing yours. The one page limit admits no exceptions, since a single page fits on a screen and is still readable. If you have figures or charts or anything else, it will all have to fit within the one page. Here is what your proposal should contain:

• Your name(s).
• A project title.
• A brief description of what you plan to do. Your classmates should be able to understand your description without knowing special things about your project. Avoid jargon. A few clear paragraphs will do.
• Pointers to the relevant bibliography. This may be a chapter out of a book or a single journal paper if you plan, say, to implement an existing algorithm.
• The form in which the final result of your work will be delivered. If it is a program, state clearly what the input is (images in which format, and where you are going to obtain them), and how the output is going to be visualized (another image, a list of numbers, a plot, a display in an interactive tool, etc.). Also, state what programming language will you use (you are free to choose any system convenient to you). If you compare programs, state your criteria for comparison, and so forth. If it is something else, say what the report will look like and how you will test your results, if applicable.
• A schedule of work, including what you plan to submit for the project review. List the tasks ahead of you, and split the available time in reasonable proportions and among the people in your group. This is a very important part of the proposal. The point of the proposal is to show that you plan to do something reasonable given the time, not something overly ambitious. Your judgment in this part will be considered for the grade. Do not worry if your proposal sounds "small": there will be plenty of opportunities to get stuck and for your schedule to slip, so do not bite off more than you can chew. In my experience, the best projects have had a modest target but are carried out with care and attention to details. [ God is in the details. - Mies Van Der Rohe.]
• Anything else that helps understand what you plan to do and establish the feasibility of the project in the limited amount of time that is available. A picture that illustrates the project may be worth many words.

Project Review

The aim of the review checkpoint is to make sure that you do not wait for the end of the quarter to start working. For the review, you should try to meet the goal you stated in the proposal. For instance, if you proposed to implement an algorithm, most of the code should be ready at review time. Minor bugs are still possible, and the input-output interfaces may be incomplete, but the core parts of the main algorithm should be shown to work. If you plan to modify existing code, the original code should be up and running, and most of the modifications should be complete. The review is really the substantial deadline for this problem. Do not hand in the actual code, just an HTML page that shows inputs, outputs, discussion, and a link to the source code.

Final Report

After the review, you should plan to spend the last few days writing up a final report, running experiments where applicable, and cleaning up the details. You should not plan any major coding for this time period. Writing clearly requires time, and the clarity of your final document will be a substantial part of your grade. Again, a small investigation carried out with care and insight is much better than an ambitious project executed in a hurry. The final paper should be in the format of a publishable paper. You should not necessarily aim for publication, since you may just be reproducing known results, but your paper should be as well-organized and clearly presented as a research journal paper. It should contain a good, brief introduction, figures where appropriate, and a sensible conclusion outlining what you have learned from your work. Again, your final report must be an HTML document, with as many links and pointers as you need.

Html

HTML is a simple markup language for World-Wide Web pages. Here is a pointer to tutorials, reference manuals, and tips for writing in HTML. However, the easiest way to learn how to write in this language is to copy an example and modify it to suit your needs.

And now, off to work, and have fun!

Back to the CPS296.1 Home Page