I did a little test setting the threading at different amounts and testing
the rotation controls in the SphericalTransform node on Windows and it
appears that the higher I set the threads the slower the performance is,
almost increasing exponentially.  The default setting on the workstation
I'm on at the moment (Win 7, 2x Intel E5-2689, 64gb) is 32 threads.
Threads at 1 or 2 performs like butter.  Might be worth having the devs
investigate why there's a performance hit with this node, must be a
threading issue.

Best,
John

On Thu, Mar 28, 2013 at 2:21 PM, Michael Garrett <[email protected]>
wrote:

> Hey Gary,
>
> If I'm understanding yo correctly, one way you could abstract out this
> rotation to make it a bit more clear is to project the cubic environment
> through a sixpack (cubic camera) setup and then you can just parent the
> setup to an Axis to do the master rotation.
>
> Check this nice tutorial Frank R did a while back for some pointers on a
> latlong>cubic>projection workflow.
>
> http://www.youtube.com/watch?v=Uq0kcUJ_XA8
>
>
>
>
> On 26 March 2013 11:03, Gary Jaeger <[email protected]> wrote:
>
>> Can someone give me a sanity check on what to expect with the
>> SphericalTransform node? I'm doing something I *think* should work fine,
>> but isn't.
>>
>> Imagine we want to take projectors and project onto the ceiling, walls
>> and floor of a room. Pretend our room is a cube, and we can get the
>> projectors into position such that their 90 deg FOV maps perfectly to
>> walls, ceiling and floor.
>>
>> Our approach has been to render a vray 360 cube and crop it 6x (or just
>> render individual cameras) , run those through SphericalTransform nodes to
>> break out the -x, +x etc and then pipe those into a separated cube mesh to
>> "check our work" as it were. We enter values (-90, 180, etc) in the output
>> rotation fields to spin the -x to the LEFT etc. All good so far, and things
>> look correct. The sides are all in their correct places. The trouble comes
>> when we want to rotate that cubic SphericalTransform to spin the content on
>> the surface of our cube mesh.
>>
>> It seems like we should be able to rotate the output rotation values and
>> have the imagery spin on the "walls". We expression tied the output
>> rotations together (we also tried the input rotations) so we could "spin"
>> all 6 together. It "works" in the sense that the imagery spins, but the
>> imagery gets all split up. It feels like maybe a rotation order thing, or
>> we just doing something fundamentally wrong.
>>
>> We *should* be able to do this, right?
>>
>> Gary Jaeger // Core Studio
>> 249 Princeton Avenue
>> Half Moon Bay, CA 94019
>> 650 728 7060
>> http://corestudio.com
>>
>>
>> _______________________________________________
>> Nuke-users mailing list
>> [email protected], http://forums.thefoundry.co.uk/
>> http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users
>>
>
>
> _______________________________________________
> Nuke-users mailing list
> [email protected], http://forums.thefoundry.co.uk/
> http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users
>
_______________________________________________
Nuke-users mailing list
[email protected], http://forums.thefoundry.co.uk/
http://support.thefoundry.co.uk/cgi-bin/mailman/listinfo/nuke-users

Reply via email to