Cineon Log transform problem

[Nicolas Hanson]

I'm setting up the basics on a music video with the following structure as the video is shot linear.

1. Linear to Cineon Log 
2. New node with a film emulation and at the same time putting it back to 709

When transforming the materiale to Cineon it creates some strange chrome artifacts on her lips. I have not managed to "roll it back" when working on the first node. It looks like it's created in the transformation. Any tips on how to deal with this?

Original shot

Invert

Final grade with lip problem

 

[Andy Minuth]

By 'Linear' I assume You mean a videoish space like Rec.709 / 1886.

The problem is that there is no perfect conversion from a low dynamic range color space like Rec.709 to a high dynamic range space like Cineon-Log. Going the usual way from Log to Rec.709 some information is lost, especially in the very bright and dark and also highly saturated areas. When you try to invert this lossy process, usually artefacts appear. 

This is a LogC to Rec.1886 Curve that I plotted years ago (for Legal and Extended Range which is unimportant here). Pay attention to the flat parts of the curve in the shadows and highlights. When you invert a curve like this, these parts are very steep and these code values are getting stretched a lot. This means that a tiny difference in the original results in a big difference in the result.

Here I put two luma waveform plots of different Video (Rec.1886) to Cineon-Log transforms. You can see that the first one is very smooth compared to the other one. The first one is practically free of artefacts, while the second one has some problems. This is just for the luma part of the transform, additional problems occur for the colours (but this is more difficult to visualise). 

Filmlight optimised and smoothed the inverse transforms that are used by Truelight Colour Spaces within Baselight, to produce less artefacts. @Daniele Siragusano talks about it in this video around 06:20 - 06:50

 

After the explanation, my suggestions how to solve it:

• Generally I suggest to avoid a workflow like this (Video -> Log -> Video). But I know that usually the colourist cannot influence the camera settings  
• Try to use a different transform, that produces less artefacts. I don't know if there are choices in Resolve. In Baselight the cleanest one is the inverse transform of the 'Truelight Video 1' DRT.
• Generally You can reduce the artefacts, when You avoid the extreme ends of the transform. This means it will probably work better if you reduce the contrast of the image, before you send it through the conversion. Once the picture is converted to Log without artefacts, You can increase the contrast as much as You want. If You have highly saturated colours in the picture, You can additionally try to reduce the saturation a bit before the conversion.

[Paul Dore]

If the transforms are being performed using LUTs then that might explain some of the issues you're getting. A 'cleaner' process may well produce better results.

[Andy Minuth]

Out of curiosity I put Your example into Baselight and converted it to a Log space. All the available transforms are free of artefacts. Here is one result. Additionally I noticed that in Your inverted picture a lot of noise gets pushed, which is not optimal I think.

 

[Paul Dore]

There should be no artefacts or information loss if the transforms are performed mathematically in 32bit float, which I believe is the case with Baselight colour management as well as RCM in Resolve. Only when there is an analogue emulation (film stock) is a 3D LUT utilised, but other than that all transforms are based on linearise/de-linearise formulae and 3x3 matrices, so in theory are all invertible without loss of information.

[Andy Minuth]

You are right, Paul. But the formula/matrix based transforms are also more "boring" and less popular among colourists. In general I suggest to use whatever transform one likes to go from Log to a display space. But for the inverse I advise to choose a clean one, that is not based on film or includes gamut de-compression. 

[Paul Dore]

Yes, why let actual precision and colour accuracy get in the way of a more "fun" and popular approach. Not all colourists think that way though. Some are more technically minded (heaven forbid), in so much that they actually know what they're doing, whereas others are more like colour plumbers, with a toolbox of tricks and a gut instinct. Fair enough, the world will always need plumbers. However, a technical problem thats requires precise mathematical formulae cannot be solved with just positive attitude and excellent client skills, and trying to bluff your way around that won't improve the quality of your images either.

I suggest perhaps trying to raise the bar a little (instead of lowering it even further), and putting the onus on enhanced technical knowledge rather than what best suits your mood or whim.   

[Marc Wielage]

I wouldn't try to push it back into log. I would start with what the filmmakers shot with and take it from there, and make sure they understand they shot themselves in the foot by shooting in Rec709 in the first place. It limits the available choices and they just have to live with that. Be aware that hundreds, maybe thousands, of some of the most memorable videotape videos of all time were shot in Rec601 (standard def) or Rec709 (HD). It's not rocket science, it just takes time and effort to make it work. 

I would also tell the client to expect less. Larry David's philosophy is good: "Curb Your Enthusiasm." 

[Nicolas Hanson]

Thank you very much guys, I ended up working with the linear images. I used a combination of curves followed by lift, gamma gain to get closer to the rich contrast that was baked into the film emulation LUT. It's a shame that the invert LUT doesn't work any better than this because I like to work this way.

[Daniele Siragusano]

if you use the Truelight Video 1 inverse DRT, this is based on formulas rather than inverse LUTs. It gives normally a much smoother Result if you want to go from display referred to scene referred. This will a very common problem if you have stock or archive footage inside a project that needs HDR Deliveries.

Also inverse Transforms cannot be avoided in certain situations (for example if you are forced to record film out)

Simple Formula Matrix Colour Space conversions are used to convert from one space to the other without change the image state really. (scene to scene or display to display space conversion (within one viewing condition)), if you want to change the image state more complex transformations are needed. @Nicolas Hanson Which conversion did you used in the first place to produce the artefacts in your examples?

[Nicolas Hanson]

Thank you for the great explanation @Daniele Siragusano. I don't have access to an inverse DRT, so I simply used a built in inverse LUT. I should probably head over to Baselight for tasks like this as I can't access any formula based transform in Davinci.