[Bf-committers] GSoC 2016: Finish multi-camera reconstruction

[Matthias Fauconneau]

Hi Sergey, Thomas,

I did not delve into the code yet.
Maybe first step would actually be to write an overview and some
documentation on what we already have ;)
I remember Keir also did some work on tracking for ToS, creating an
hybrid doing SAD then KLT, IIRC ?
I agree full automatic tracking might be a bit ambitious for a new
student and especially a too open ended problem.
On the other hand, a project delivering part of the solution could
already bring useful improvements.
For example, working on the detection part as you mention.

Best regards,
Matthias

On Fri, Mar 4, 2016 at 7:50 AM, Sergey Sharybin <sergey.vfx at gmail.com> wrote:
> Thomas, will be really good to have you involved :)
>
> Matthias, did you check current state of Libmv before writting the proposal?
>
> As for detectors, we had FAST and MORAVEC detectors for some time, but then
> i threw them into canal and put Harris detector instead. It's not opimized,
> but after some work could be made to work in auitomatic pipeline.
>
> As for implementing autotrack as a GSoC project, afraid it's a bit too
> ambitious to sepect something like that complex from a single student
> working fo few months. Would suggest being really careful before submiting
> to such a task.
>
> On Thu, Mar 3, 2016 at 11:53 PM, Matthias Fauconneau <
> matthias.fauconneau at gmail.com> wrote:
>
>> Hi Thomas,
>>
>> During GSoC11, I developed 3 detectors and an API which Sergey used to
>> expose feature detection in Blender.
>> I do not know why it did not make the final cut, maybe it was not
>> polished enough yet to be useful for production.
>> I think, at this point, Sergey knows best the latest state of the
>> motion tracking features and would be the one most able to sort out
>> all your ideas into useful projects.
>> It is not trivial to estimate beforehand how much work is involved in
>> specific features, but should the main project (multiview) shorter
>> than expected, I would be open to complete with sides.
>> The main priority would still be to first make sure the new tool is
>> properly integrated, known and tested.
>> Especially since there will be no VFX open project to finish things off.
>> As Ton says, since we have the benefit of having the users we develop
>> for around, there is certainly some room to guide the direction of the
>> work on the project.
>>
>> Best regards,
>> Matthias
>>
>> On Thu, Mar 3, 2016 at 7:13 PM, Thomas Beck <software at plasmasolutions.de>
>> wrote:
>> > Hi Matthias,
>> >
>> > as we use the Blender camera tracker pretty often in our projects, we'll
>> > follow this GSOC closely. One feature that we were really missing
>> > (especially in long shots with many easily distinguishable tracking
>> points)
>> > is the possibility to automatically track our footage. It should work in
>> a
>> > similar fashion as "throw in markers every frame automatically so you
>> get a
>> > good coverage, track those until they get lost, erase the obviously
>> > erroneous markers, solve it". We talked with Keir at Bconf 2014 and he
>> said
>> > there is smth. was planned by Sergey and him. So maybe this could be a
>> > (side-) project as well...
>> >
>> > Best regards,
>> > Thomas
>> >
>> >
>> >
>> > Plasmasolutions
>> > Design | Development | Training
>> >
>> > Website: Http://www.plasmasolutions.de <http://www.plasmasolutions.de/>
>> > Blog: Http://blog.plasmasolutions.de <http://blog.plasmasolutions.de/>
>> >
>> > Telefon: +49 176 2017 9565
>> >
>> > Blender Foundation Certified Trainer
>> > <http://www.blendernetwork.org/BFCT>Autor von "Blender 2.7 - das
>> umfassende
>> > Handbuch" <http://www.galileo-press.de/3404>
>> >
>> > 2016-03-03 16:57 GMT+01:00 Jed <jedfrechette at gmail.com>:
>> >
>> >> Matthias Fauconneau wrote
>> >> > I am not sure how the interest in Blender VFX is these days, maybe
>> >> > there are some other features users are looking forward too.
>> >>
>> >> I haven't been following development closely so this may already have
>> been
>> >> implemented, but last time I checked one of the big features missing
>> from
>> >> Blender's tracker was the ability to use known 3D points to constrain
>> its
>> >> solves. There's a good chance productions using witness cameras to
>> improve
>> >> their solves will also be doing set surveys, whether that's with a tape
>> >> measure, total station, or lidar; so the ability to utilize that
>> geometry
>> >> data would be very valuable.
>> >>
>> >>
>> >>
>> >>
>> >> --
>> >> View this message in context:
>> >>
>> http://blender.45788.x6.nabble.com/GSoC-2016-Finish-multi-camera-reconstruction-tp147565p147619.html
>> >> Sent from the Bf-committers mailing list archive at Nabble.com.
>> >> _______________________________________________
>> >> Bf-committers mailing list
>> >> Bf-committers at blender.org
>> >> http://lists.blender.org/mailman/listinfo/bf-committers
>> >>
>> >>
>> > _______________________________________________
>> > Bf-committers mailing list
>> > Bf-committers at blender.org
>> > http://lists.blender.org/mailman/listinfo/bf-committers
>> _______________________________________________
>> Bf-committers mailing list
>> Bf-committers at blender.org
>> http://lists.blender.org/mailman/listinfo/bf-committers
>>
>
>
>
> --
> With best regards, Sergey Sharybin
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> http://lists.blender.org/mailman/listinfo/bf-committers