Black Levels Quicktime (uncompressed/Animation) vs REDCine-X

[Gavin Greenwalt]

I've read through dozens of REDUser threads on this issue and everyone seems to be investigating the completely wrong culprits in all of them. Jumping to the finish line... can REDCine-X just not export Uncompressed or Animation quicktimes that match the viewer? I've had infinite trouble with Quicktime destroying my colors, but never in Uncompressed or Animation. Those are my two safe places. Are even those problematic in REDCineX? I can always render off frame sequences and make a quicktime from those, but it adds an annoying step.

Windows 7
REDCine-X Pro Beta 8

 

[Gavin Greenwalt]

Nevermind. EXRs are borked too. (Checker of Nuke EXR vs REDCineX). But slightly borked in a different way (EXRs are lighter blacks and greens).

 

[Gavin Greenwalt]

Good news is that TIFF sequences do come out correct (Checker of Nuke TIFF vs REDCineX Viewer).

 

[Gavin Greenwalt]

Ok, now things are getting interesting.

Nuke R3D reader vs REDCineX EXR. Identical. So the viewer and TIFF writer evidently are on a different color pipeline than the EXR writer. Which is annoying since an EXR should look exactly the same just with overbrights no? REDCine-X should be doing all of the color processing in 32 bit floating point right? So what extra color operation is the Viewer in RCX-Pro doing that it's not applying to EXRs but is applying to TIFFs?

 

[Gavin Greenwalt]

Ok one last comparison. Note the excess contrast in the EXR checkers. Note the greenish tint to the skin tones as opposed to the nice natural skin tones from RCX/Tiff output. Note the increased noise from the excess lifting in the EXR as well.

This is exactly the same in the Nuke R3D reader as the RCX EXR output. The RCX TIFF output is way "better".

 

[David Ibbitson]

Gavin,

When doing EXR's we ignore most of the color settings, and the Gamma space is forced to Linear.

Specifically, we ignore:

- RGB gains
- Contrast
- Brightness
- Gamma Curve
- Curve, Luma, R,G, & B
- Shadow
- Lift Gamma Gain R,G, & B

The RCX viewer is doing no special color operations. However you see 8 bit RGB in the viewer, whereas TIFF and EXR exports are 16 bit. I'll
have to try the Quicktime uncompressed/animation and see if I am getting the same difference.

Are you viewing your quicktimes in 7 or X?

David

 

[Graeme Nattress]

EXRs are necessarily linear light, and how they're being viewed in Nuke is via a display LUT. That is why they look as they do. There is, however, a good correspondence between EXR, and REDLogFilm imported and interpreted as cineon to linear light - then they're identical.

Graeme

 

[Nick Shaw]

Ok, now things are getting interesting.

Nuke R3D reader vs REDCineX EXR. Identical. So the viewer and TIFF writer evidently are on a different color pipeline than the EXR writer.

Nuke's half-float linear R3D decoder is the same as REDCINE-X's EXR pipeline, so should give the same results. If you are using REDgamma(2) in REDCINE-X it will only match a Nuke decode using the same gamma curve. You need to bear in mind that REDCINE-X's EXR exporter (and Nuke's half-float linear reader) map the sensor black level to float zero, whereas the integer pipeline leaves the sensor black level elevated (as recorded) to preserve the noise floor, and requires the user to use e.g. a shadow adjustment to sit the black level down as required.

I totally understand why this discrepancy exists, and RED's reasons for doing it this way, but it does make matching the results of an integer pipeline and float pipeline rather complex. REDlogFilm get around this by mapping sensor black to Cineon black (10-bit code value 95) which is why I am particularly fond of it when using an integer (preferably 16-bit) pipeline.

 

[Gavin Greenwalt]

Gavin,

When doing EXR's we ignore most of the color settings, and the Gamma space is forced to Linear.

Specifically, we ignore:

- RGB gains
- Contrast
- Brightness
- Gamma Curve
- Curve, Luma, R,G, & B
- Shadow
- Lift Gamma Gain R,G, & B

The RCX viewer is doing no special color operations. However you see 8 bit RGB in the viewer, whereas TIFF and EXR exports are 16 bit. I'll
have to try the Quicktime uncompressed/animation and see if I am getting the same difference.

Are you viewing your quicktimes in 7 or X?

David

When you say "forced to linear" do you mean redgamma2 isn't applied? Is it not possible to linearize with a 1/2.2 gamma?


So are you applying lift etc in gamma corrected space not linear space is that why it's not being applied?

 

[Graeme Nattress]

Gavin, we always ensure EXRs are utterly linear to light. That means nothing that would cause a deviation from utter linearity gets applied. So indeed, REDGamma2 is not applied.

Graeme

 

[Gavin Greenwalt]

Thanks. I think got all the Linear/Half-Float Linear/EXR/DPX/REDLogFilm/sRGB/REDGamma business straightened out in my head with a good battery of tests.

Were you able to replicate the Uncompressed/Animation QT shifts though?

 

[Nick Shaw]

Were you able to replicate the Uncompressed/Animation QT shifts though?

For me Animation compressed QuickTime and TIFF sequence from REDCINE-X match perfectly in Nuke. I'm on Mac, so can't speak for what happens under Windows. To ensure they match I have to make sure that both are interpreted as the same "colorspace" (although I don't agree with that term – it's really "linearisation transfer function") in the Nuke Read node. Nuke defaults to "Gamma 1.8" for QuickTimes, and "sRGB" for TIFFs, which will make them appear different when they are not!