10 replies to this topic

[mdharrington]

I made a custom lut using this process
http://www.pigsfly.c...?showtopic=7159

What I was trying to do was have a LUT emulate the difference between a linear RED file and a REDGamma2 Red file. Quite often I have to work in Cineform and its conversion does not preserve the Red metadata values and my idea was work in a linear gamma space with a LUT applied on top


So i imported the same r3d file twice...set one to linear, the other to redGamma2....did a histogram match between the 2, then CC's the LUT Cubecreate tool with the instance of the CC, ran that through the analyzer and wrote out the LUT...

Now when that LUT is applied through the view LUT or the file LUT the image is way off....not even remotely close
any ideas why this would be?
Comp attached

Attached Files

•  LUT test.comp   7.56K   3 downloads

[ChadCapeland]

I'm going out on a limb and saying that A003_C002_1207ML_001 looks NOTHING like the output of LUTCubeCreator? The matching in CC is based on histograms, and for this technique to work the histogram of R3D file has to match that of the LUTCubeCreator, which is perfectly flat. Ignore the fact that CC is broken for sampling the LUT cube image that doesn't cover the whole rectangle.

[mdharrington]

So there is no hope in this technique?

The idea seemed sound...histogram match auto corrects the linear r3d to the gamma encoded r3d.....the same curve is applied to the LUTCube...then read by the LUTanalyser.
Does the difference in image size change the histogram?

[ChadCapeland]

The technique is sound, but the tools are not up to the task. The CC can't expose the transform for another set of inputs.

Maybe you could try matching with CCv instead?

[mdharrington]

I will give it a shot,
I noticed the histogram match was close....but not 100%, might be something for a custom tool
thanks for the help

[ChadCapeland]

The CCv match works, but either way, your footage sample has to have very good coverage of the gamut you wish to correct. Maybe take representative frames over the whole sequence and quilt them together to generate your sampling?

[mdharrington]

ya that did the trick....much better,
thanks Chad....

still not 100%...

is there any benefit to having a lut cube size 66 or bigger?
and is there a higher precision form of LUT from any of the choices in the LUT analyser (probably not, but ya never know)

[ChadCapeland]

The larger the LUT cube, the more high frequency the change that can be captured. So for a size of 66, you're ONLY able to capture a change that occurs over 1/66 of the 0-1 range. To put it another way, you couldn't represent any change that you wouldn't be able to do (in 1D) with a CCv with 66 evenly spaced points on the curve, with those points all being set to linear. It's not bad but it's not good if you are reading out colors to more than 2 decimal places.

The LUT analyser isn't really needed. Just using the LUT cube images, you can do everything. The only reason to use the LUT analyser files is when you need to send the LUT to another application that doesn't know how to use LUT images.

[mdharrington]

thats actually my main purpose....to send out a lut to another application....basically I am making a LUT to apply to cineform metadata to emulate reds custom gamma curves, and can integrate cineform vfx with redcode for editing, I assume the 3d LUT interpolates through the values it dosn't have stored...I guess it would take a 1024 cube to fully capture 10 bit then, which would be massive

btw i noticed a bug in my a/b split view on my main fusion screen the b side is always slightly darker....even when comparing identical footage (dvi monitor quadro card) window....when i put the view on a third monitor with an a/b split the LUT looks very close.... imperceptible by eye and could only see a difference using the difference matte...

[bfloch]

mdharrington, on 16 May 2012 - 10:27 PM, said:

btw i noticed a bug in my a/b split view on my main fusion screen the b side is always slightly darker....even when comparing identical footage (dvi monitor quadro card) window....when i put the view on a third monitor with an a/b split the LUT looks very close.... imperceptible by eye and could only see a difference using the difference matte...

Are you sure that you have exactly the same settings for both A and B? It sounds very unlikely that results differ.

[mdharrington]

yup your right...

it appears there was some kinda refresh issue on the normalize button....the comp was very unresponsive in the viewport to changes but when i open a new one everything works out fine
my bad....cancel the re-write