Computational Photography

I am taking the course right now (7th week fall 2024). I selected the summer 2024 due to the limitation on this site. The quality is very low. The lectures have low value when you compare the time instructor put into them and off course you put in to get/learn something out of them. Too much guideline to let themself free of guilt. This course technically ruined the image I had of Gatech. Their care is using OpenAI, They don't understand that people use OpenAI when they do not understand the course and face a task that is totally new to them. There is no challenge/learning when you feel helpless in an assignment. I wish I could get refund.

I took Computer Vision previously. In that course (run by the same professor, thus worth the comparison), although there was a steep learning curve, I learned a lot of math and interesting implementations. The lectures were too lengthy with some bad jokes here and there, but they took time to actually explain technical stuff clearly. The report writing was straightforward and on point, taking me little time to finish once I had the results.

For this course (Computer Photography), I privately hoped it was sort of breathing room for me after CV. Well, in terms of difficulty of the material/assignments, that might be true, but it was not in terms of shear workload and uncertainties they intentionally created for you to have some anxieties. In my opinion, the whole experience is not worth the effort if one has already taken CV.

1. The video lectures are not very helpful because they are light on explaining math/algorithms clearly, but heavy on demonstrating the published results. 2. The ideas/techniques implemented in assignments/projects are not hard, but a couple of them require careful book-keeping, which are dull to me. 3. The reports took significant amount of effort to write due to broad questions but rigid grading rubrics. The requirements for discussion in these reports feel like beating a dead horse. There's algorithm discussion, then there's code analysis step by step down to the specific line of code or function. Maybe it was meant to deal with some perfunctory students historically, the problem prompts were often written in a rather condescending tone and try to coerce every bit of things you have to say about the subject. This encourages students to write long but mumbling essays on their not-so-straightforward experiences of doing the assignment. 4. There are a lot of research papers/links/challenges to worry about if you strictly follow their instructions, but in the end, your understanding of those materials is only touched lightly.

Overall, they seem to want you to emerge from fighting against a lot of busy work. Maybe that's a good starting point for vision-related topics later on?

I took Computational Photography after I took Computer Vision- while I can't necessarily recommend anybody else doing this, CP makes CV look like the worst class I'll have taken in my entire time in OMSCS.

I'm projecting a high B/low A prior to final grades coming back.

For this course, the assignments were mostly straightforward, the projects required some paper reading to re-create research results, and there were a few Jupyter Notebook/Canvas quizzes. Peer feedback on reports for assignments was, predictably, useless and often not useful for constructive feedback on assignments. Final exam is open note/resource/etc.

I never felt overwhelmed in this course, the TA's answered questions without being overly Socratic on Ed, and the lectures were easy to follow and well made by Essa.

Definitely recommend taking this course, especially if you are pursuing the CP&R specialization like me.

Grade achieved: A

Course was incredibly difficult. It's constituent parts are not bad. The lectures, the projects, tests, reports, autograder but the cumulative effect makes this course overwhelming.

If you hit the requirements and manage to understand how to effectively navigate the autograder you will be graded fairly.

ATTEND THE OFFICE HOURS LECTURES. This was critical to getting algorithm hints for implementing the white papers.

The most intimidating part of the course was having to read and implement very formal academics papers with math you may be unfamiliar with reading. I will say as infuriating implementing the papers was the boost in skills and confidence to do this type of implementation is satisfying once the course is over.

Once of the worst course I've done during my life.

Super lazy TA's always mentioning they have life and other work to do. If you don't have a time and should constantly cry about it, just don't be TA.

No recurring office hours.

Very boring homework. Everything is super strict.

• Function names, arguments.
• If something is wrong by 10^-9, it's marked as wrong.
• Limited submissions, that don't increase even when it's a mistake on TA's side. I think, they do their best to hide their mistakes
• Can not import simple things

Professor and lectures look fine. Though, I think, course would benefit from major refresh in TAs and homeworks

The class overall was difficult but very rewarding. I thought each assignment did a good job of relating back to the lectures in the course.

Overall 5 assignments which varied in difficulty. Some were much easier than others but each assignment came with detailed docstrings on the logic that needed to be implemented. The instructors were willing to help through ED as people reached out as well

There were 2 projects which were the hardest portion of the class in my opinion. These provided no clear instructions other than implementing an algorithm directly from a published paper and asking questions through ED. There were about 2-3 weeks allotted for these projects and you definitely needed at least a full week of work to create something meaningful.

The final exam as open everything and drew a lot of content directly from the lectures and the assigned reading that supplemented the lectures towards the end of the course.

This has been my 6th class in the program and overall probably my favorite. I've also taken HCI, SAD, AI4T, AI and ML4T. This had a similar difficulty to AI in my opinion.

This course is less linear algebra intense than Computer Vision but has a lot of overlap both in code and lectures in the beginning. Prof. Bobicks lectures are more math heavy while Prof. Essa's are more conceptual, and since I've already taken Computer Vision, I did not find it necessary to watch the lectures for this course except for some later lessons when taking the final exam (2x speed). If you feel you are struggling during this course, I would recommend you watch Bobicks lectures, esp around Laplacian pyramids and homography.

However, the course is more time consuming than Computer Vision because of the addition of reports and the fact that after the 2nd assignment, with the exception of an additional week for each project, you will have something big due every week. This contributes the majority of the stress of the class if you're trying to get an A, as assignments past the first 2 count 10% and projects count 15%. Similarly, peer feedback and the notebook quizzes are weighted heavily enough that they matter. Getting a B, however, is much more achievable and with less stress.

Some overall positives:

I do like the fact everything you do creates interesting results / outputs that you can then show off.

I thought that this class was well run, esp thanks to Bob and Kimberly.

I also really enjoyed both Projects where you implement algorithms from research papers. If you struggle here be sure to brush up on vectorization in order to debug much more quickly and definitely attend office hours.

Overall though, I am neutral / liked just because of the early overlap with CV, and the amount of stress and time consumed. I think Essa's lectures really shine in the later lessons and wish there was more focus on this.

This course was much harder than it should be. Glad I dropped to preserve my emotional health; this course isn't worth the stress and work. But I'm sad because I actually enjoy the subject matter and wanted to learn more about it.

Here are some specific criticisms. Note that I dropped after the second project, so things may have changed later on (although I can't imagine things got significantly better).

• The professor of record (Dr. Essa) was non-existent -- didn't even say hello to the class. The TAs run the show.

• Lectures were mostly surface-level discussions and oftentimes irrelevant. Sometimes interesting, but I didn't like the lecture style.

• Assignments were fair. Bits of coding in Jupyter notebooks.

• Projects were awful and the reason why I dropped.

  • The TAs offered no help. Every TA answer to a question was a variation of this:

    Please reread the project instructions -- all the hints and help you need are there.

    Sounds fair, except that those "hints" were sometimes cryptic and were spread out in any of 10 different files/places:

    • some were in the project announcement/"help" posts on Ed
    • some were scattered throughout the project files
    • several were random comments spread throughout the 5+ files of the LaTeX report template

  • There are no office hours. Not even ones recorded from previous semesters were made available.

  • You are penalized for checking code against the auto-grader -- points are deducted after X submissions. Other courses use velocity thresholds (X submissions allowed every Y hours), so not sure why that's not implemented here (should be a OMSCS standard at this point).

  • They are very strict about no code sharing, yet they are extremely specific about what your functions need to return -- and because the projects are basically one long pipeline, simple errors in an early function are devastating.

  • But you can't share any code, even the most simple test cases -- they are very strict about that. Yet the tests they provide are useless.

  • So you are left to code your own, with no collaboration, no help from the TAs, no way to check against the auto-grader, no useful lectures.