[Bf-committers] Bf-committers Digest, Vol 152, Issue 20
levonh at gmail.com
Sat Mar 25 16:54:01 CET 2017
> From: Tianwei Shen <shentianweipku at gmail.com>
> Subject: Re: [Bf-committers] GSoC 2017: Camera breathing support
> To: bf-blender developers <bf-committers at blender.org>
> Message-ID: <14FAAA4C-AD72-4C9B-969C-BF6C4A90165D at gmail.com>
> Content-Type: text/plain; charset=utf-8
> Hi Levon,
> Thank you so much this long reply. First of all, I?ve been looking for
> user tests and suggestions for the multi-view reconstruction project. If
> you have ideas for making it better, just feel free to drop me emails. On
> the other hand, it is still a quite large patch. So we need time to split
> it up and gradually merge it into the master. But hopefully this can be
> integrated well with the camera breathing support project and even
> automatic tracking in the future.
> As for the camera breathing support, I didn?t realize lens distortion
> parameters would also change with the focal lengths. I thought we?d only
> deal with changing focal lengths with the zoom-in/out motions. So things
> become complicated here since it seems to me that focal lengths and
> distortion parameters cannot be estimated on the fly. Users have to first
> calculate this information (focal lengths for each frames and their
> corresponding lens distortion parameters) using some calibration tools. Can
> the solver reliably deal with changing focal lengths and distortions? On
> the other hand, if users have to first calculate focal distances using some
> tools (if Blender doesn?t have its own) in the first place, would it impose
> burden and inconvenience for users?
technically its possible to counteract lens breathing with just lens
distortion, by uniformly scaling the calculated distortion grid, however
its not something blender is able to do, and i havent seen it done in any
other software, it was just an idea we were playing around with at some
stage at work.
cool, i will take a few days to write up some more info and try and break
it down to specific ideas you could work on... Im happy to help 'mentor'
your GSOC in a way, but my coding skills are not up to scratch and im only
familiar with parts of the blender code, so would still require some extra
feel free to ask away any other questions.
Jacob Merrill wrote:
what about using a object of known scale to calibrate (like a 3d printed
yes, we do this all the time, alot of film sets have lidar scans of the
entire set, that we can use to help the solver.
having known accurate 3d data makes determining a change in focal length,
focus distance, and therefor lens distortion way easier and less manual
work. its still quite a complicated workflow, but it allows the computer to
do more of the work, and less trial and error by the user.
More information about the Bf-committers