Re: Introducing generation of non-fixed size filters/kernels

From: stefan (stefan_at_[hidden])
Date: 2019-07-12 13:57:35

On 2019-07-12 4:58 a.m., Olzhas Zhumabek wrote:
> Hi,
>
> I thought about adding filter generation to GIL, and it is needed for 
> Harris corner detector code. The problem is that they are currently 
> inside my Harris corner detector PR, and I thought may be they are 
> worth extracting into a separate PR.

Good idea !

>
> Here is the list of currently implemented filter generators:
>
> 1. Unnormalized mean (all entries are 1)
> 2. Normalized mean (all entries are 1 / (w*h))
> 3. Gaussian blur kernel (https://en.wikipedia.org/wiki/Gaussian_blur)

As always, please coordinate with Miral to make sure you are not both 
working on the same algorithm (or at least, if you do, do it in a 
coordinated way :-) ).

>
> I can cherry pick commits into new branch, the problem is that the 
> changes are mixed with Harris corner detector code.

Moving self-contained algorithms out of your current PR seems fine, in 
particular if they are accompanied by their own docs, tests, etc.

Thanks,

Stefan

--
       ...ich hab' noch einen Koffer in Berlin...