Other DRTs as real LMTs in ACES 1.3

priikone · July 18, 2021, 9:20pm

As an exercise to learn more I wanted to see how easy it would be to make current ACES version look like some other DRT such as the popular ARRI K1S1 by creating an LMT. And I wanted to try to do it with just a tone curve and a 3x3 matrix, and not do something more and turn it into a 3D LUT (and definitely not doing the full invert thing).

And to my surprise, it worked pretty well. Not nearly perfect, but considering how simple they are, I’d say not too shabby. For normal surface colors they give decent match but less so in higher luminance levels and with super saturated colors. I did compare them against Baselight’s equivalent scene looks (3D LUTs) through TCAMv2 and they are visibly better, but the LMTs are not too far off. (I may do 3D LUT LMTs later for better match as another exercise.)

I put together a repo for the LMTs. There’s LMTs for ARRI ALF-2, ARRI K1S1, RED IPP2 and SONY S-Gamut3.Cine, plus some other stuff. All LMTs are available as CLFs (v3.0) and I also generated DCTLs for Davinci Resolve.

Some images:

I think the per-channel DRT does half the work to help with the match, but the RRT also brought some issues. The biggest obstacle to getting a better match was the red-modifier in the RRT. It affects skin tones significantly and I found that I had to push the red more to overcome that effect (I prioritized skin tones). This then meant that other colors went off a little bit more, and made also yellows more saturated (3x3 matrix after all is very limited). I tried also without the red-modifier in the RRT and I found that it was easier to get a better match compared to what I got in the end, for whatever it’s worth.

I hope somebody finds these useful or interesting. And let me know if you spot anything amiss with the LMTs. I was thinking of doing AMF variants as well, but I have no idea how to test AMF at the moment.

A side note: I did want to do this same also for OpenDRT but that didn’t go anywhere. It became immediately clear that it’s not possible to get a match with just a tone-curve and 3x3 matrix. The LMT would have to do more and be a 3D LUT in this case, or some additional rendering transform would have to follow the simple LMT. But, I will continue playing with the OpenDRT.

priikone · September 21, 2021, 8:21pm

I did a second round of DRT LMTs. This time the LMTs use 3D LUT to get much closer match to the original DRT. For ordinary surface colors the match is now pretty much exact.

I also decided to split the LMTs into three categories. There’s now the emulation LMTs which emulate the DRTs as closely as possible, but I also added a “colorimetric LMT” which changes scene colorimetry but doesn’t do anything else (I hope the name make sense). They also retain saturation over the luminance range while the DRT emulation LMTs don’t. And a third category is Tone-curve LMTs. All of the LMTs have at least 22 stops of dynamic range (0-1.1 in ACEScct).

In my mind splitting the looks like this makes sense and provides user many more possible combinations for new looks.

And here’s some images for comparison:

These looks were applied with ocioconvert using the CLF versions of the LMTs. More examples images in the repo…

I hope again somebody finds these useful. I might add some new looks later, but I have no plans to add more emulation LMTs.

meleshkevich · September 25, 2021, 5:21pm

Looks great! Thanks for sharing this!
There are some sharp transitions with LUT stress test images, but it’s probably outside of the useful gamut.

Could you please tell more about how you created them? Sounds like a miracle to get almost a perfect match just with 3x3 matrix.

Also I found it only works from default LUT folder. Resolve supports custom LUT folders now, but seems like DCTL can’t use 3D LUTs when it’s outside of default LUT folder.

priikone · September 25, 2021, 6:36pm

Thanks for testing and this comment. The “colorimetric LMTs” don’t compress saturation at all so they have all the same issues with super saturated colors as current ACES has in general and the Reference Gamut Compression can help with those.

For the DRT emulation LMTs I did expand the gamut (quite linearly really) and they do transition to more saturated colors than the original DRTs. This was my choice, I didn’t have to do it, but I wanted more saturated colors available, and I wasn’t concerned about getting exact match with super saturated colors (it wasn’t going to be possible anyhow with the wider dynamic range). The transitions (to super saturated range) is quick but it should be smooth.

BTW, did you look at the original DRT LUTs through LUT stress test? Some are better than others…

No no, these don’t use 3x3 matrix. That was the first experiment. These new ones are constructed 3D LUTs (and 1D LUTs for tone curves).

meleshkevich · September 25, 2021, 6:44pm

Could you please tell where to find these? Resolve has ACES 0.1.1 emulation LUT. And also ACES 1.2 has one built-in DRT (which is only one, and isn’t a LUT, so you are probably talking about something different). I’m not a technically skilled person, well, I probably understand a bit more about color science, than usually colorists do, but compared to all of you guys, I’m an absolute newbie. I’m just a colorist using ACES. So my knowledge is very limited.

And thanks for the explanation!

priikone · September 25, 2021, 7:01pm

Specifically for the emulated DRTs, these are the sources for the originals:

For ARRI ALF2 and K1S1: LUT Generator | ARRI
For Sony: LUT & ART Files | Sony Cine (scroll to the end of the page)
For RED: Downloads Details

And also from Filmlight’s page provide most of them: FilmLight | Support | Customer Login | Truelight Colour Spaces

I might have taken the RED LUTs from the Filmlight’s page, if memory serves.

I’m newbie too! These LMTs were the most complicated thing I’ve done with 3D LUTs… ACES Central is amazing place to learn. There’s so much information buried in the pages and so much talent present in the forum.

meleshkevich · September 25, 2021, 7:29pm

Oh! Now I get it! I only see the word DRT here on the forum, so I thought you’re talking about previous ACES DRT, and I totally forgot that K1S1, IPP2 LUTs and whatever else are also Display Rendering Transforms.
ARRI LUTs for rec709 are artifact free for me. But PQ LUT from their generator have something strange happening between magenta and red colors. Also their DCI-P3 LUT have D65 white point for some reason.
IPP2 LUTs are also accurate for me. But what I don’t like about them is that they place 18% gray to 463 or so in 10 bit full levels. It’s too bright. With rec709 it’s 418. And most of other DRT place it even slightly below that.

Derek · September 20, 2022, 6:04am

I’ve been experimenting with these DRTs and wanted to share something I stumbled upon.

Here’s the light sabers image in ACES with all the familiar issues

And here it is with the RED IPP2 Look.

Here’s a couple examples from the gamut compression test images

Notice how the blue goes to magenta in ACES on these, but does not in the IPP2 LMT

Happy accidents!

priikone · September 20, 2022, 8:05am

That LMT compresses gamut which explains the differences. They do not however deal with negative input values so they can still cause issues. For negative input values use RGC before the LMT.