This discussion seminar will explore the state of the art in computer graphics, focusing on topics relevant to our local research, mainly through reading papers in the graphics research literature.
This course teaches the skill of reading research papers, as well as the graphics topics that the papers are about. That means learning to approach difficult technical material that you will generally not understand completely; part of the trick of reading research literature is to be able to learn something from the fraction of a paper you understand, without getting lost in the part that you don't. Even accomplished researchers don't understand everything in most papers they read; it's usually only when you are going to implement or build on a particular paper that you put in the many hours needed to study it completely.
That said, it takes substantial background knowledge to understand enough from these papers that it is worth your time to read them. The official prerequisite is “graduate work in computer graphics or vision” which is a proxy for having enough general knowledge of math and physics and enough specific familiarity with computer graphics. In practice the course is designed for the needs of PhD students in graphics, but I expect PhD students in vision, robotics, scientific computing, and nearby fields will be fine, as well as advanced undergraduates and Master's students who enjoy challenging math and have done coursework in graphics.
The weekly process for this class is:
You should expect to spend at least 3 hours a week on this process: an hour each reading papers and an hour in class. But you will definitely get more out by putting more in!
I set the reading list (it's not a democracy like reading groups often are) but will definitely respond to participants' interests. The readings will not always be planned out that far in advance, and selections may depend on discussions that we have about earlier papers.
||Just one paper this week in honor of Fall Break.|
A key part of this paper's contribution is the optimization method that can be “interpreted as casting differentiable rendering into the framework of Sobolev preconditioned gradient descent,” in the authors' words. This idea shows up in more geometry- and simulation-oriented graphics papers such as:
This is an S/U class, and the expectations are that you will read the papers and participate in the discussion. The practical way to measure this is by noting that you are posting questions about the papers. So the requirement to pass the course is to post questions almost every week (it's no big deal to miss one here or there).