Colour artefacts or breakup using ACES

[quote=“Joey_Foreman, post:34, topic:520”]
Thanks so much, Nick. Finally got around to trying the LUT, but having a strange issue. Pretty sure I placed it in the correct location[/quote]
Odd. Sorry about that. I’ve never seen that error message. I just downloaded and checked the downloaded LUT, and it works for me. Although I am running v12.5.2 on the MacBook Pro I am in front of right now.

It looks like the right LUT folder. You can double check by clicking the Open LUT Folder button in the Color Management tab of Project Settings.

Also, just in case the LUT got somehow corrupted in the download, try this zipped version:

2 Likes

Hi everyone,
this is my very first post in this community, unfortunately i won’t be able to give back as much as i would want, i know close to nothing about colour science and coding.
I am manly a lurker, trying to read and learn as much as i can, I love what aces is about and i actually love to grade in aces, i have done many projects in it.

Anyway, I started testing for my next project and i got the same artefact you were talking about in this thread.
I have tried to use the FixHilights DCTL that Scott Dyer posted before and even if it makes it slightly better it doesn’t fix the problem.

I attach an example of the artefact that i am getting and i will link a dpx of the flat image.

I also want to take the chance to thank everyone that i contributing so much in this community, it’s really helpful!
Orash

Davinci Resolve 12.5
ACEScct
PRORES444XQ
https://drive.google.com/open?id=0B-kmzfWwMPPUQlBseGZITWxOS0E

Interesting example, and thanks for sharing the frame.

What is the encoding of the DPX? In other words, what camera did you select for your Input Transform (IDT)? Was it Alexa?

I assumed Alexa, processed using ACES w/ and w/o the the matrix adjustment and got the following results:

Is this similar to what you saw when applying the DCTL? (If not, then it’s likely something was not applied correctly…)

1 Like

Hi Scott!
thanks for the quick reply.
Yes, it’s alexa, I have applied the alexa idt as project base and srgb (as i am delivering for web) as odt.
I’ve then applied the dctl on the first node, the result was slightly better but not as good as your one.

Could it be that by transcoding into dpx 10 bit i have cut out some information on the super saturated pixels?
My test were done with prores444xq straight from camera, i know is not ideal as they are not raw, but aren’t them 16bit?

Edit: today i will try with thee rec709 odt and let you know how it goes, maybe that’s the problem.

Alright,
I have done few more tests and i didn’t get any success, neither switching odt nor using the dpx rather then the PR444XQ.
I will post the screenshots of my settings, hopefully i will learn something new and really good :smile:

That’s probably why. The DCTL I shared was designed to be applied to linear ACES data (ACES2065-1). To correctly apply it, right click on the clip in the media pool or, if on Color page, right click on the clip. About 3/4 down the menu (just below where it lets you set a per-clip ACES Input Transform) there are options for No LUT, 1D LUT, 3D LUT, or DaVinci CTL: Try applying the DCTL to any problematic shots using this DCTL menu instead of the regular nodes.

The reason yours was not working correctly is because DCTL applied via this per-clip contextual menu is applied directly to the ACES image. Any DCTL applied in the node tree is applied to the current working space - such as ACEScct, in your case - unless additional “conversion” DCTL is used to get the data to the correct state for the application of the matrix.

For example, if you instead chained together a sequence of ACEScct-to-ACES → LMT.Academy.FixHighlightImageArtifacts → ACES-to-ACEScct in your node tree then you could get the same visual result as applying DCTL via the contextual menu. Hope this makes sense and that you can similar results to me. Good luck!

Just wanted to let you know that I have used the LMT Fix Highlight yesterday on a project that was all shot with super saturated color gelled light.
Basically every shot had massive artefacts everywhere, not just highlights.

The LMT worked like a beauty, fixed every artefact and made every shot a lot more pleasing to work, I didn’t have to fight saturation at all.

I will ask permission to post some before and after!

1 Like

Yes that LMT works like a beauty. Really appreciate having it also. I usually apply it to all shots. As sometimes it’ll fix some weird bokeh that might go unnoticed.

2 Likes

@Orash_Rahnema @chuckyboilo glad to hear it has worked well for both of you thus far!

1 Like

Hi, I would like to share my footage on the same issue.

Origial RED R3D : DaVinci ACES ; ODT = Rec.709

Origial RED R3D : DaVinci ACES ; ODT = Rec.709, with “LMT.Academy.FixHighlights.dctl”

ACES AP0 in LogFilmScale ( No RRT+ODT ) just want to see image data after IDT.

DaVinci YRGB , REDWideGamutRGB + Log3G10 then apply default AlexaLogC_709 LUT.

And here is the original (trimmed) r3d file from camara.
https://drive.google.com/file/d/0B2nuWtWHwLhlaXJFbHhmZWRqWHM/view?usp=sharing

I’m still trying to fix this artfactes.

You are putting the LMT in the wrong place. You probably applied it via a node.
This Highlight-Fix-LMT must be applied to the clip via the contextmenu of the clip. NOT in the node tree.

I just checked and it works fine.

Peter

3 Likes

Thanks Peter! Now it’s completely fixed.:heart_eyes:

I just did another test.
I upgraded DaVinci Resolve to V.14 then apply my own LUT on a node and It surprisingly works!

The LUT is simply a matrix transformation from standard AlexaLogC to AlexaLogC with blue primary moved to rec.709.

I put it here if someone want to try.
https://drive.google.com/file/d/0B2nuWtWHwLhlcmxHSXQ4djZHWUk/view?usp=sharing

Not sure if this is the same issue exactly, but I am experiencing something similar in blue parts of a CG render. Below is the render as it appears in Nuke with the ACES 1.0.3 OCIO config.

When it is brightened, say with a gain, the area around the highlight on the blue sphere acts quite oddly:

I tried to apply the matrix provided by Scott in this thread by adding it to the OCIO config. Perhaps I am doing that incorrectly. I’m assuming the matrix should go at the end of the input color space for ACEScg:

  • !
    name: role_rendering
    family: Utility/Aliases
    equalitygroup: “”
    bitdepth: 32f
    description: |
    The ACEScg color space

    ACES Transform ID : ACEScsc.ACEScg_to_ACES
    isdata: false
    allocation: lg2
    allocationvars: [-8, 5, 0.00390625]

to_reference: ! {src: ACES - ACEScg, dst: ACES - ACES2065-1}

to_reference: !<GroupTransform>
  children:
    - !<ColorSpaceTransform> {src: ACES - ACEScg, dst: ACES - ACES2065-1}
    - !<MatrixTransform> {matrix: [0.9404372683, -0.0183068787, 0.0778696104, 0, 0.0083786969, 0.8286599939, 0.1629613092, 0, 0.0005471261 -0.0008833746 1.0003362486,, 0, 0, 0, 0, 1]}

At any rate I could not get it to work. Any help is appreciated!

The correction matrix should be applied to ACES2065-1 (linear/AP0).

Not 100% sure if that’s what you’re doing here, but that’s the space in which the matrix is intended to be applied.

It looks like you’re bringing in the spheres as ACEScg, converting to ACES2065-1 and then applying the matrix?

Also, just curious what is the colorimetry of the spheres? What red, green, blue do they represent?

Yes, is that correct?

Not quite sure what you mean with colormetry. Sorry that’s a bit over my head. They are rendered in Vray and the colors are just randomly picked. The blue is <0.96, 0.391, 1.0>

Here’s the EXR file if you’d like to try it out.

I don’t think this is the same issue as the others in this thread. I would say that what you are seeing is exactly what I would expect from this image.

If you look at the Nuke colour picker reading in the area surrounding the specular highlight on the blue ball, you will see it is ACEScg[0.097, 0.754, 5.473]. This has a much higher blue value than the equivalents for the other balls. The Rec.709 and sRGB ODTs clip above about 16.29, and when applying gain, 5.473 will clearly hit this quite soon (at gain more than 3) causing that area on the blue ball to flatten and desaturate, which is exactly what I see.