worst experience yet in OMSCS. The course content is good, the professor is good, but the TAs are very irresponsive and you would never get timely help or suggestions from them. Only take the class if you are sure you have enough time to struggle with meaningless guess and trials (of the project questions). BTW, while the projects are not well described, TAs will never show mercy when taking off your scores.

I'm presently enrolled in Computer Vision in Spring 2022. We're now working on the last assignment, assignment 6. I am tracking towards an A but in order to achieved that I've had to spend more than 30 hours a week and often take days off work to do homework--which is what I'm doing today as I write this review.

This course is completely brutal in its workload, especially for those who work full-time. Someone posted on Slack the other day that this course is like having a second fulltime job. On February 13 I started keeping track of the enrollment numbers by counting the people's tab on Canvas. Between February 13 and the drop deadline of March 16, 20% of the students dropped the class. Given that some students had likely already dropped, the drop rate after the first week or two is likely at least 25%. The brutality of the course would be forgivable if it felt like it was rewarding, but it does not. A significant part of the workload is in trying to tweak already correct algorithm implementations so that they pass the autograder. Autograding often has specific unspecified requirements that you have to discover on your own through trial an error. It is not unusual to make 25 autograder submissions in a single assignment as you employ trial and error.

Another large part of the workload is in customizing weak algorithms to solve a particular instance of a hard problem (for example, tracking a particular object in a particular video with occlusion). This customization does not generalize, and so the hours you spend customizing an algorithm to the particular challenge is largely useless. The workload of this course is substantially disproportionate to its learning value.

Unlike most other courses, as other reviewers have noted, the TAs provide nearly no help. They'll answer questions about technical requirements of an assignment, but provide almost no help to students who are having difficulty. They usually do not respond to student questions at all, but when they do I have not observed them being in any way rude or condescending as some other reviewers have suggested. However, though the TAs are not helpful, the course lecture videos from Prof. Bobick are very good. So students can resolve most conceptual issues on their own by reviewing the videos. So, it doesn't seem like having more responsive TA's is critical to improving the course. The main problem is just the tremendous workload that doesn't doesn't result a comparable level of learning.

Like many other recent reviews, I recommend students avoid this course. Take note of the fact that considering the 20 most recent OMSCentral reviews, including this one, the mean and median hour requirements for this course are 28.7 and 30 hours. In retrospect, I expect I'll view this course like a long illness or trauma that substantially interfered with my life but with limited redeeming value.

08/05/2022 Update:

I think PS2, PS3 and Final Project are the most difficult assignments in this course. Anyway, it is not hard to achieve a high A if you can spend 25-35 hours like a second full-time job.

Pros

• Good lectures covering traditional CV methods.
• Fair Grading. TAs only focus on the requirements, especially for image results. They graded the final project generously. At least better than 21Fall DL TAs.

Cons

• Some content overlapped with Computational Photography. Lack of new DL methods.
• TAs are less active and cannot reply questions in time. In most of time, they only provide the assignment criteria.
• Prof absent.

22/02/2022, 12:23:51 AM

This course taught traditional CV. Good course content.