Re: Tetrahedral is Best

Jeremy Selan <jeremy...@...>

So if you look in the aces configuration, you'll see there are two .ocio files.  config.ocio and config_1_0_3.ocio.   config_1_0_3 explicitly specifies tetrahedral interpolation for the rrt 3dluts.  The reason we havent chosen to roll that into the normal config.ocio is that tetrahedral interpolation was only added in 1.0.3, and I dont want to leave those folks locked off on prior versions (such as nuke 6.3v8) with configs that dont load.    (If anyone has a better idea on how to approach this, please let me know).

But this does raise the question, of what should OCIO::INTERP_BEST do?  Currently it defaults to linear for both 1d and 3d luts.  It would be a one line change to default BEST to tetrahedral for the 3dlut case.

In peoples experience, are there ever cases where trilinear is preferable to tetrahedral for 3d lut interpolation?

In terms of GPU tetrahedral support, we have some ideas how to add it in, and hope to explore them by the end of summer.

-- Jeremy

On Tue, May 8, 2012 at 9:54 AM, Brendan Bolles <bre...@...> wrote:
I was experimenting with the ACES configuration the other day. I
applied the RRT to a grey ramp and saw some subtle color artifacts,
which could be verified by boosting the saturation. I tried applying
the rrt_ut33_sRGB.spi3d LUT directly and saw the same artifacts, until
I switched to tetrahedral interpolation which made them go away. Maybe
there's something to this tetrahedral business after all. ;)

So maybe should we switch ACES and other configurations to use
Tetrahedral when doing a 3D LUT?

Also, one of the interpolation types is "Best", which currently uses
Linear for both 1D and 3D LUTs. Seems to me the best interpolation for
a 3D LUT is Tetrahedral. Ben says it's actually faster than tri-linear
in his tests, although I don't think anyone's ever figured out how to
do it in the GPU.


Join to automatically receive all group messages.