[libcamera-devel] ipa: rkisp1: Add autofocus algorithm

[Daniel Semkowicz]

Hello all,

First of all, please forgive my lack of communication. The last year
was very intense for me, and I missed some messages in the mess...
I also had to stop the work on the AF for a longer period.

I see that there have been attempts to continue my work.
Was there any progress in the area of auto-focus on rkisp1?

I was also contacted by Luis Garcia, who tried to work on the AF
for a while. He made initial implementation of golden search instead
of hill climb method. However, the main code base of the AF series
is the same. He shared his work here:

  https://github.com/luigi311/libcamera/tree/golden_search_focus

Unfortunately, Luis cannot continue the work any time soon.

I should have more time in the near future, so I can support anyone who
is working on the AF series. If no one is working on it, I may resume
this series.

Kind regards
Daniel


On Mon, Aug 28, 2023 at 1:45 PM Jacopo Mondi
<jacopo.mondi at ideasonboard.com> wrote:
>
> Daniel now cc-ed for real
>
> On Mon, Aug 28, 2023 at 09:07:12AM +0200, Jacopo Mondi via libcamera-devel wrote:
> > Hi Nicholas,
> >    sorry for not replying on irc but I was on leave
> >
> > On Tue, Aug 22, 2023 at 09:37:07PM -0400, Nicholas Roth via libcamera-devel wrote:
> > > To whom it may concern,
> > >
> > > I've had my eye on this patch series for some months now, and can see
> > > that it has not progressed since March. I am interested in
> > > contributing. My understanding is that the reason this ISP lacks an
> > > autofocus implementation is that the algorithm here does not work
> > > properly because it gets stuck in local minima. Is that correct?
> > >
> > > Professionally, I am a research-adjacent AI practitioner and I work
> > > with and read a lot about optimization, which seems very closely
> > > related to autofocus. I have also done computer vision work in the
> > > past. This is why I think that I might have something to add here.
> > >
> > > If this sounds good to the maintainers, and the original author
> > > doesn't object, I would like to proceed with this understanding,
> > > though I can't guarantee a fast rate of progress. Please also let me
> > > know if I am missing any other context about the patch series that
> > > would likely be helpful.
> >
> > I've now cc-ed Daniel (the original author).
> >
> > I admit I have not followed after Laurent's review of v6, maybe Daniel
> > can give more details about the current status.
> >
> > Thanks
> >   j
> >
> > >
> > > Thanks,
> > > -Nicholas