Re: Feature Detection

From: Olzhas Zhumabek (anonymous.from.applecity_at_[hidden])
Date: 2019-06-08 00:49:37

>
> However, I see on the above link that there are multiple (at least three)
> Free (as in Software) implementation of A-KAZE, so I expect there to be no
> problem.
>
Yeah, probably we can use it.

 How did you expect to use the code ? Incorporate it directly (by cloning)
> ? Or referring to it as a third-party library user would have to install
> first ?

I was thinking about using its output as goal, or at least rough goal. I
did not expect to extract any parts of it, may be have a look if something
in the paper is ambiguous.

 I think it would be great if users could switch between different
> implementations. In fact, it might even be interesting to establish a
> common API, then provide wrappers for the three above, and use benchmarking
> to measure how their performance compares. Or perhaps that's too ambitious
> and we need to restrict ourselves to one implementation.

The main obstacle as I see it is that most implementations pander to opencv
2.x or 3.x. And since they are mostly C-style codebases, it might take
considerable effort to change the interface. I believe the main issue is to
get layout-compatible views, and the rest should be easy. I believe I
should focus on items listed in my gsoc proposal, and afterwards we could
try that. In fact, I have a senior project coming up next study year, so I
have personal interest in doing it.

Olzhas