Academic problem solving in Computer Vision

Question

I am currently pursuing a Master degree of Engineering and I will have to write a thesis about solving a specific Machine Vision application. I have done some CV projects in the past but now I am tinkering about the methods of problem solving in the broad field of object detection. What I noticed is that it looks like there are about two kind of people trying to solve CV problems and asking about them on a SE site.

The first kind of people needs to solve a specific problem (like, finding faces or horses or else in a picture) and is asking questions like "Oh, I need to find an algorithm to detect face" or "How can I detect cars with SIFT and OPENCV".

The second kind of people is writing academic papers on some new algorithms (like SIFT, ORB, etc.). They might have questions like "How to best compute the Hamming distance" or else.

Both kind of people and questions are relevant and fine to me (if my opinion matters). I wonder though if there is somewhere I could ask questions or share thoughts about more structured methods of computer vision problem solving. I mean, I would like to know a place where I could ask questions such as:

"What are the general steps of solving an object detection (or recognition) in a 'graduate' manner, and being able to demonstrate that all the steps used where necessary and to a high confidence the best method was chosen?"

I wanted to post on Signal Processing Meta but you need 5 rep in order to...

BTW, the question asked is a real one that I would like to ask here, simply it might not be the place or respect the guidelines of this SE site. In the same way, I feel this might be too CV related to ask on Academia SE...

Edit
What I mean by "a graduate manner" is the rigorous approach to problem solving that one should use in the course of graduate studies. What I mean is that sometime (especially when you work in a company vs working in academia), you must rapidly get a solution that work to solve a problem, instead of doing an entire review of the possible methods and doing benchmarks and etc... (This is a generalization and I don't want to imply that working in a company makes you use somewhat botched methodology, see it more like a "Perfect is the enemy of good".)

Answer 1

With all due respect, your question is rather vague... For example, it is not clear to me what you mean by "structured methods of computer vision problem solving", or what you mean by a "graduate manner" of object detection or recognition.

If you are doing a Master's thesis in computer vision then presumably you have an adviser, who is supposed to be knowledgeable in this field, and who is supposed to help you define your problem, and point you to the relevant papers.

By the way, defining and constraining your problem is the key here. "general steps of solving an object detection (or recognition)" is not a very well defined problem. Are you trying to detect a specific object, such as a particular book or your favorite coffee mug? Or are you trying to detect members of broad categories of objects, such as faces, cars, people, or chairs?

Does your method need to be invariant to in-plane rotation? The answer is probably "no" for faces, which are rarely upside down, but "yes" for detecting cars in aerial images. There are many more questions like this.

So, my point is that object detection, let alone recognition, is a broad area, and it is best to talk to your adviser about scoping it down.