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

[Matthias Fauconneau]

Hi,

Answers are inlined.

On Thu, Mar 3, 2016 at 6:48 AM, Sergey Sharybin <sergey.vfx at gmail.com> wrote:
> They are related indeed, and in fact should be combined with some other
> projects since they aren't that big actually at all.

Alright, I was not clear that items on this page are not necessarily
standalone projects.

> Where exactly it's done? To my knowledge we only have implementation based
> on OpenCV. That is unacceptable for Blender. Calibration is to be redone to
> use Libmv's basement and support it's distortion models and such.

Had to do some digging here ;)

1)
It seems I first did an OpenCV checkerboard calibration tool (
libmv/contrib/opencv/calibration.cpp ).
This could be ported to be standalone if needed.

2)
Towards the end of my GSoC I did a standalone tool which estimates the
distortion from straight lines.
It leverages a single file code ( algebraic_lens_distortion.c ), I
remember we clarified any licensing issues.
Otherwise, it could also be replaced by a simple numerical optimization.
The tool was requiring the user to input the lines but could also
easily be modified to detect them.
As it was the end of GSoC, it seems the code was never integrated with
a corresponding Blender frontend.
https://gitorious.org/qt-lens-distortion/qt-lens-distortion

> There is quite a big interests in VFX along Blender's users.

Hopefully this will translate to good user feedback and test cases ;)

I think it would make the most sense for me to work on multi-camera
reconstruction since I already have the experience from GSoC11 and
developing libmv's nView tool (3D reconstruction from multiple views).
Not sure how that would fit with Tianwei though, since he is also
interested in this project.
Maybe we could figure out more work to do and/or a way to split the SfM work.

P.S: I was wondering what has been done during GSoC13 ?

Regards,
Matthias