Re: [ggt-devel] Problems with Quats and scales on Matrix44f

[Doug M. <mc...@ia...>]

kevin meinert wrote:
> I think when there is a scale in the matrix, that the quat you get is not
> a rotation quat.  a rotation quat by definition is unit length == 1. 
> having the scale in the matrix no longer means you're encoding pure
> rotation, now you've got something more, rotation plus scale.  translation
> works because it's a completely separate part of the matrix, and so isn't
> read during the conversion.  the scale however (as i'm sure you know)
> exists in the same values that encode rotation (the upper left 3x3
> coordinate axes).
> 
> separating scale from the rotation in a matrix that encodes both is
> slightly ambiguous, and apparently not supported by the well known
> mat->quat algorithms found in the books and websites we used as reference
> to write this code.   I've never seen a variation on this algorithm,
> though that doesn't mean it's not doable....   for now if you do want
> scale you'll have to keep your scale matrix separated out from your other
> transforms... (basically get your rotation quat _before_ you compose your
> scale matrix into your matrix transform).
> 
Thanks Kevin. That is what I discovered and the solution that I 
implemented as well. Everything works great once I grab scale out of the 
matrix before creating the quat. And just to verify, when grabbing scale 
out of a 4x4 you take the length of each of the first 3 column vectors? 
Thanks again.

Doug

> *--*---*---*----*-----*------*------*-----*----*---*---*--*
> Kevin Meinert                         /_/
> http://www.subatomicglue.com                    \        /
>                                                \/  __    \/
>                                                    \__
>                                             \_\
> 
> 
>>Hello,
>>     I am using gmtl 0.4.9. If I try to create a quat from this matrix I
>>get the following:
>>
>>| 0.5 -0.866025 0 0 |
>>| 0.866025 0.5 0 0 |
>>| 0 0 1 0 |
>>| 0 0 0 1 |
>>
>> quat
>>(0, 0, 0.5, 0.866025)
>>
>>Which gives me the correct end rotation. Now, if I take the same matrix
>>from above and apply a scale of 3 I get the following:
>>
>>| 1.5 -2.59808 0 0 |
>>| 2.59808 1.5 0 0 |
>>| 0 0 3 0 |
>>| 0 0 0 1 |
>>
>> quat
>>(0, 0, 0.981981, 1.32288)
>>
>>Which ends up over rotating by what appears to be a few degrees. It is my
>>understanding that I should be able to create a quat with a matrix that
>>does have scale on it, correct?
>>
>>Thanks for the help.
>>
>>Doug
>>
>>
>>