Understanding and Critiquing Generative Computer Vision
School of Computer Science, Carnegie Mellon University
Course Description
This is a seminar course on generative models in computer vision. We will meet weekly and present papers, critique papers, listen to talks, and work on a project. Some classes will involve invited talks from different researchers, and others will only have student paper presentations and discussions. The overall aim of the course is to delve in-depth into the nuances of modern day generative models, understand how they work so well, and develop reasonable critiques of their capabilities to understand how they can be made even better and applicable for diverse downstream research themes.
Logistics
When: Fridays 12:30 pm to 2:50 pm ET
Where: NSH 3002
Instructor: Abhinav Gupta
TA: Homanga Bharadhwaj
Deliverables
The evaluation will be based on the following criteria:
Paper presentation (20%): Each student will sign-up to present one paper during the course
Paper summary (10%): Each student will sign-up to lead the discussion of one paper during the course. They will initiate discussion of the paper by the prior Monday on Piazza through a summary post. So, each paper will have a presenter, and a discussion lead.
Paper discussions (15%): Each student must participate in the discussion of at-least two papers. They will post comments on the summary posted by the discussion lead. The discussions should not summarize the summary but try to develop reasonable critiques and merits of the paper.
Project (45%): Each student must contribute to a project (group size 1 to 2). We will have a project idea lighting pitch early in the semester, and a final project presentation. Each project group must submit a 2-page writeup on the project. The projects could be on anything related to generative computer vision and should involve some implementation - so paper summaries are out of scope, but model inference without any model training is in-scope.
Class Participation (10%): Students should participate in the class discussions as much as possible.
Piazza
We will be using Piazza for class discussions and updates. Students will be automatically enrolled based on class registration. Here is a tutorial about Piazza for anyone unfamiliar with it https://support.piazza.com/support/solutions/articles/48001079546-piazza-intro-for-students