Colour mismatch between LUT box and in-camera Look

[calebstaines]

Hi,

I'm trying to get to the bottom of a colour discrepancy I'm seeing when applying a LUT to the camera's Log SDI feed using an IS-mini LUT box controlled by Pomfort Livegrade. The output of the LUT box doesn't quite match the image output by the camera when using the same LUT in-camera. The LUT used in the LUT box was made in Resolve and is a concatenation of a Log-to-Log CMT LUT and Arri's LogC4-to-BT.709 DRT. The look used in-camera is that same CMT inside an ALF with the DRT in-camera set to SDR 709.

The IS-mini's output is slightly desaturated compared to the camera's output and there is a slight luma difference. It may be a deficiency in the LUT box's 26^3 processing but I'm also wondering if it's a colour space mismatch when converting from R'G'B' to Y'CbCr and back again.

Has anyone seen anything like this before using an IS-mini with the Alexa 35?

And can anyone tell me what luma coefficients are used in-camera to produce the Y'CbCr values sent over SDI when the output is set to LogC4? Are they BT.709 or BT.2020?

Thanks,
Caleb
DIT
Auckland, New Zealand

[Simon Duschl]

Hi Caleb,

are you using Tetrahedral for the LUT processing in the Resolve Studio?
-> project settings -> Color Management -> Lookup Tables -> 3D lookup table interpolation = Tetrahedral?
Maybe the LUT processing is the problem here?

In the meantime I can check for the coefficients when LogC4 is used in-camera.
Depending on the selection in the SDI Color Color Space menu, the correct one is being used (BT. 709 or BT. 2020). When LogC4 is active it's greyed out. I will check this and come back to you.

[Simon Duschl]

Hey Caleb,

for LogC4 we are always using the BT. 709 coefficients.

[calebstaines]

Hi Simon,

Thanks for the suggestion. I will check with the colourist what his exact process was in creating and exporting our LUTs. I can see that switching to tetrahedral interpolation in Resolve definitely reduces the differences I see between the CMT + DRT vs the combined LUT I was given so it's worth exploring further.

I've since compared the output of the LUT from the IS-Mini vs the same LUT through an AJA FS-HDR and the FS-HDR's output matches the camera's much more closely. So I'm leaning towards thinking that the IS-Mini's 26^3 processsing and triangular interpolation isn't accurate enough for close scrutiny. Perhaps this is more noticeable with LogC4 than with LogC3 due to the lower code value for the midtone and perhaps more aggressive tone-mapping that requires?

It's worth mentioning that I'm seeing the same difference between the in-camera look and IS-Mini's with ARRI's standard LogC4 to BT.709 LUT.

Thanks also for finding out about the luma coefficients. That's good to know for certain.

Thanks,
Caleb