[Bf-committers] Movie distortion algorithm

[j.bakker at atmind.nl]

Hi, 

Thanks for the feedback Jonathan & Michael. You are probably right.

Currently we already have a look-up grid that is precalculated. It is a
nice-to-have feature to speed-up this precalculation process. Current
lookup grid is per pixel. In a 4k shot it just takes a very long time.

Perhaps it is an idea that these grids are eventually stored to disk.

Regards, 
Jeroen Bakker.

Original Message:
-----------------
From:  mgschwan at blensor.org
Date: Tue, 17 Apr 2012 11:06:17 +0200
To: bf-committers at blender.org
Subject: Re: [Bf-committers] Movie distortion algorithm


I second that,

using a precomputed mapping will be the most useful algorithm. Because 
normally, everybody is doing the reverse mapping (undistortion by 
calculating the distorted positions from the undistorted ones and doing 
a reverse mapping) and avoids the other way around like the devil runs 
from the holy water.

What could be a feasible solution too:

Calculate the distorted points from the undistorted ones and 
interpolate between the supporting points with a 2nd order polynomial. 
Maybe you even could reduce the number of supporting points by 
calculating only a sparse mapping between undistorted and distorted 
points. Basically doing a piecewise polynomial interpolation.

Michael

Am 17.04.2012 00:43, schrieb Jonathan Merritt:
> Reading that paper (Vass and Perlaki), they present a very good 
> solution.
>  If you're running the inverse transformation on a sequence of 
> frames, all
> of which have the same distortion parameters, then just compute the 
> inverse
> transformation once ("slow"), and cache the result.  Then do a 
> look-up
> using either exact values or some kind of fast interpolation for all 
> of
> your movie frames.
>
> They also describe using Newton's method to compute the (xd, yd) => 
> (x,y)
> transformation, as I suggested.  Since these guys have already looked 
> at
> this problem, I suspect that the algebraic inverse transformation is
> probably very long and painful.
>
> Even if your distortion parameters are varying, you could also speed 
> up the
> process by using the solution from the previous frame as the staring 
> point
> for your Newton iteration of the next frame.
>
> On Mon, Apr 16, 2012 at 4:06 PM, Dan Eicher <dan at trollwerks.org> 
> wrote:
>>
>>> These guys describe a fast preview mode (if that's the problem):
>>>
>>> http://www.vassg.hu/pdf/vass_gg_2003_lo.pdf
>>>
>>> ...not that I understand the math behind it.
>>>
>>> Dan
>>> _______________________________________________
>>> 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

--------------------------------------------------------------------
mail2web.com - Microsoft® Exchange solutions from a leading provider -
http://link.mail2web.com/Business/Exchange