COMP9517 - Computer Vision

Difficulty: 2/10 · Time Commitment: 2/10 · Enjoyability: 4/10 · Mark: 94

Summary

This course is an introduction to basic computer vision concepts such as image detection and classification. It teaches a variety of algorithms that modify images for specific purposes. it also briefly covers different machine learning methods and how they are applied to different computer vision tasks. There is a lot of potential for this course and the lecture content is very interesting, however, in itself the course is very basic and most assessible tasks are very superficial.

Positives

• Great lecture content
• Very straightforward course on a topic with lots of potential

Negatives

• Really dry assessments
• None of the content is applied well and the coding problems are extremely boring and not challenging
• You will not need or have opportunities for any creativity in this course

In-depth

Structure

When I did this course, it consisted of 4 short coding labs, one short coding assignment (about double the size of a lab), a very manageable group project consisting of 5 students, and a final exam. The labs and assignments in this course are all coding based and vaguely related to the content taught in lectures. The group project is almost completely unrelated to anything learnt in the course and if extremely easy if at least 3 people on the team are able to complete a moderate amount of work. The final exam is also very straightfoward as it consists of 24 hours to submit a short commentary of an existing research paper.

Lectures and Tutorials

The lectures and tutorials in this course are combined and contain a lot of interesting content. Overall, the content taught is very straigthfoward but there is quite a bit of maths. I would recommend not watching the lectures live and instead, watching the recordings afterwards and pausing at the mathematical sections to understand them better. None of the maths is overly complex and it is explained quite well so deriving the equations provided only takes a few minutes per slide. Additionally, none of the maths is assessed so it is not essential to understand everything perfectly unless you want to understand further research better.

Tutorials are definitely not important and are not an effective use of time in this course. The course forum is a great place to ask questions and I would not recommend watching the tutorials unless you are falling behind a lot.

Labs and Assignment

The labs and assignment in this course are extremely bland. Every AI course I had done before this course emphasised the phrase don’t reinvent the wheel, however, the core component of the labs is to reinvent the wheel. The most difficult part of the coding tasks is developing Python functions to do even primitive tasks - numpy arrays are used extensively but arraywise operations, the max function etc cannot be used.

The labs were not theoretically difficult and came with a lot of links to websites with further information on completing each task. There is no coding in any of the lectures which can make some of the labs a bit challenging as they are all coding-focused. That being said, almost every machine learning course I have done has focused on theory in lectures and expected students to learn the coding themselves. It is an important skill to translate theory into code and this course is a good opportunity to practice that.

The labs for this course are all fit into the first half of the term along with the assignment. This makes the workload heavy in Week 2-5 and then much easier for the rest of the course. I would recommend starting labs and the assignment early. It is hard to correct issues and debug code sometimes and there are sections with some trial and error. However, they are overall quite useful for understanding the basic concepts of computer vision.

Project

This course also has a group project consisting of 5 students working to make very simple segmentation and binary classification models. Overall, the workload for the project could be easily managed by 2-3 students and it was hard to find enough work for everyone to do in the project. The only deliverables for the project are a video presentation and a report. Both of these are very straightforward and tend to be marked leniently.

The project itself is very dry and there is no element of creativity for students. If you were expecting to make something useful in the real-world, you would be disappointed by this project. Additionally, tutors for this course are not really helpful. Questions on the forum are not well-answered and I recommend finding other students to help with understanding concepts rather than posting complex question on the forum.

Final Exam

The final exam for this course is a 24 hour take-home exam. It consists of writing a commentary on a given computer vision related research paper. The commentary is only 2 pages in length. The research paper I was given was very basic and easy to understand, although the papers are randomised and there is a possibility of getting a much more complex paper. The commentary itself is very non-technical and doesn’t require depth of knowledge of course content. Nonethelss, understanding basic course concepts is helpful in getting a high mark. To study for this, it is important to understand concepts used in the assignment like segmentation and classification, as well as different methods of each. Make sure to use technical knowledge throughout your commentary and explain every definition used clearly.