Alexa highlights have pink artifacts and scopes are wrong - Davinci Resolve

Tags: #<Tag:0x00007f964c7465f0> #<Tag:0x00007f964c746460>

(Brett Rayner) #1

Hi All,

I was trying out ACES by doing a directors cut grade in ACES instead of the usual method in Davinci Resolve. I set the IDT to ALEXA and the ODT to REC709 (It was shot on the AMIRA, but there is no IDT for that). I am getting some weird artifacts in the blown-out highlights of the sparklers as you can see in the images below. If i set the IDT to SRGB and just manually grade the picture to a REC709-ish shot then the artifacting does not occur

Also, in some of the shots, the scope is showing odd behaviour in severely clipping color channels, such as the red channel in the screenshot below.

(The compression seems to be making some strange artifacts, my grade does not do this)

I would love some advice for resolving the highlights issue, and maybe some insight as to why the scopes behave like they do.

(Nick Shaw) #2

Firstly AMIRA LogC is identical to ALEXA LogC, so you are using the correct IDT.

If you are on Resolve 14, you could try the “Neon Suppression” LMT, which is an implementation of the matrix that @sdyer posted on another thread here for removing this kind of artefact. If you are on an older Resolve version you can try the DCTL version of the same matrix posted in that thread.

As for the reason that artefact occurs, I suspect it may be a deBayer problem occurring due to the dramatic transition at that edge. If the colours being interpolated are very different, particularly if one or more channels are clipped as here, it is possible for the interpolation to produce “unreal” negative values, which can lead to this kind of artefact. This is only my speculation though.

Using the wrong (sRGB) IDT will not produce the negative values, as the10-bit code values below 95, which represent negatives in LogC, still represent positive values in sRGB.

Finally, when you say “data below the black point”, do you mean “below 64”? The ACES Rec.709 ODT black point is at zero, not 64, so there is nothing below black on output. It is clipped at zero. But saturated cyans in the image may cause the red output to clip to zero. Grading operations which reduce the cyan will lift some additional red out of clipping on the scopes. But that does not mean there was previously anything below zero on output. Does that make sense?

(Peter) #3

I cannot say what the reason for that is, but ARRI themselves has that problem listed in ther FAQs about ACES. Maybe that helps, too.

On that page search for: Are there any ‘gotchas’


(Brett Rayner) #4

Hi Nick,

Thanks again for coming back so quickly with detailed answers. I am not familiar with how to implement an LMT into Resolve 14, I have found very little info on that online, maybe I will just try the DCTL version. If you have any knowledge on how to do this I would be grateful!

Unfortunately my understanding of the math of these colour conversions are limited, so I am not too sure how positives and negatives are relevant in LogC/sRGB etc. I am looking to learn though, currently working through a FxPHD course on color spaces by Charles Poynton.

By “black point” I meant my forced black point by using the black clip function in resolve to make all levels crush out at 64, giving a raised black point feeling to the grade. In normal YRGB grading when I raise the black clip, no info can exist below this, and as such the scopes would be clear at the bottom. It just makes grading difficult as visually there are raised blacks, but on the scope I am seeing that something is still touching pure black.

I hope this makes sense!

(Brett Rayner) #5

Thanks for this, I find it interesting that the problem is invisible with a YRGB regular workflow but the ALEXA IDT makes it so apparent.

Still good to know its a documented phenomenon

(Nick Shaw) #6

The built in “Neon Suppression” LMT is already in the LUT folder in Resolve 14, and you apply it like any other 3D LUT. Best to apply it “at source” by right-clicking either in the Media Pool or on the Colour page thumbnail.

I definitely recommend Charles’ courses to learn about colour science.

You have to remember that in ACES you have the RRT and ODT applied after your grade. So even if you have nothing below 64 in the working space (ACEScc or ACEScct) the s-curve of the RRT+ODT is likely to push some values below 64. In DaVinci YRGB, nothing happens after the grade (unless you use RCM) so if the grade limits to 64, the final output is limited to 64 too.

Bear in mind also that just because one channel has some values at zero does not mean there is any pure black in the image. An RGB triple of [0.0, 1.0, 1.0] is 100% cyan, with a luma value of 79% in Rec.709. But the red value is zero.

(Brett Rayner) #7

Hi Nick,

Thank you once again! This makes complete sense! I now feel more confident about what I am seeing in resolve’s scopes.