Re-create Abstract profile linear=g1 in ART

Software
1

Hello !

I’m a student in 3D/CG, and have been using rawtherapee to process raw files for use in applications such as Nuke or Substance Painter.

I basically use RT to transform a raw file into a AcesP1 32bits tiff file (all other aspects are to be not used : no undistort, no tone mapping, nothing except a white balance).

But what is really giving me issues is the size of the tiff files : from 100Mb raw to 500Mb tiffs… Which in the long run will be a significant margin of my drive TBW.

I discovered ART can export in EXR, which is exactly what i need. (at least i believe i’ll be able to store my images much much more efficiently as stored data to size ratio).

So i tried “copy/pasting” the same settings i do in RT but in ART. I got them all, except for the “abstract profile : linear g=1”…

So would anyone know how to input this same abstract profile “effect” in ART ?

2

Its an interesting question…I thought EXR files had no colorspace and were linear and its at the stage of the viewing application not export that you would assign a display transform/colorspace etc…

I look forward to seeing the comments on this…

3

Hi @Achille ,
Selacting aces linear as output profile should do what you want

HTH

1 Like
4

I was coming back to suggest that as a possibility…

2 follow-ups…

Would it also be worth changing the wp to the same to enhance the data pass through… unless it is and I have changed it away to rec2020…

And second I have a profile that comes up just called linear…

Is that a default one that also comes with ART or have I added that in the past…

image
image442×136 4.83 KB

image
image248×196 10.7 KB

5

If you don’t touch the image, it won’t make any difference

And second I have a profile that comes up just called linear…

Nothing I recognise, sorry

1 Like
6

I thought…sorry to pester you I just didn’t want to do a fresh install just to sort out what comes with ART and what I added but I might go and clean this up anyway…

Thx as always…

7

Hi,
Thanks for your answers.
I’m already set in working space : ACESp1, and output in RTV4_ACES-P1 (copy pasted the ICC file from RT).

Here are all the operations i do on a raw file loaded by default :

  • clip-out-of-gamut colours: Disable
  • Highlight reconstruction: Balanced
  • Tone curve 1: Linear, Standard
  • Tone curve 2: Linear, Standard
  • White Balance
  • Working Proifle : ACESp1
  • Output Profile : RTv4_ACES_AP1
  • Demoisaicing : AMaZE
  • Border : 0
  • Lens profile : None

I couldn’t find/understand what “abstract profile : linear g=1” does in RT (in addition of the rest) to copy it onto ART.

I found it kind of funny that ART is less “scientific” than RT but has the options to export in .exr :smile:

Otherwise would there be any way to implement myself the exr export from RT ?

8

Are you sure the RTv4_ACES-AP1.icc profile is linear? At the least the version I have in the RT installation has (non-linear) tone curves defined. Below is a snip of the output from the IccToXml.exe program (it converts ICC profile files to XML).

<redTRCTag>
<parametricCurveType>
<ParametricCurve FunctionType="4"> 2.399993896484 0.947860717773 0.052139282227 0.077392578125 0.039291381836 0.000000000000 0.000000000000 </ParametricCurve>
</parametricCurveType>
</redTRCTag>
<greenTRCTag>
<parametricCurveType>
<ParametricCurve FunctionType="4"> 2.399993896484 0.947860717773 0.052139282227 0.077392578125 0.039291381836 0.000000000000 0.000000000000 </ParametricCurve>
</parametricCurveType>
</greenTRCTag>
<blueTRCTag>
<parametricCurveType>
<ParametricCurve FunctionType="4"> 2.399993896484 0.947860717773 0.052139282227 0.077392578125 0.039291381836 0.000000000000 0.000000000000 </ParametricCurve>
</parametricCurveType>
</blueTRCTag>

RT has an ICC profile creation tool (file browser tab, upper right corner, triangle with a plus sign). You could create your own profile with the AP1 primaries and select linear tone curve for it.
Harri

(The IccToXml program is probably from here: Releases · InternationalColorConsortium/DemoIccMAX · GitHub)

2 Likes
9

This topic reminded me some past discussions about RawTherapee. I found this one, which may help: Creating/getting an ACES AP0 LIN profile/PP3 to 16bits TIF - #6 by XavAL

Elle Stone produced a huge number of well-behaved ICC profiles, there are some ACES profiles with gamma=1:

2 Likes
10

Thanks for your guidance Harri and Sébastien !

I loaded as output profile “ACEScg-elle-V4-g10” from elle’s profiles.
Now the 32bits .tiff match between RT and ART ! (apart from highlights, probably due to the “highlight reconstruction method” that differs between the 2).

But when i export in .exr (PIZ or DWA compression), they have a warmer tone !

Here is the 2 images merged in “difference” blend mode in nuke :

image
image1715×1141 145 KB

You can see the overall red tint, and white spots that are the highlights not matching.

Would you have any idea why it is doing that red tint ? And how could i get a better highlight reconstruction in ART (in RT it’s “luminance recovery” algorithm)

11

They are quite different… Just to be sure I’ve understood correctly: the difference is between a TIFF and EXR export, both exported from ART?

12

Yes exactly. .tiff 32 bits and PIZ/DWAB exr (both compression yield a difference in warmer tone compared to the tiff)

13

I tried reproducing what you are doing as close and possible… I don’t have Nuke but I’m comparing the images in Python: and yes, I’m also getting slightly different images, TIFF compared to EXR. I think @agriggio 's help is needed here to figure out what is going on…

1 Like
14

Could you be seeing something like this, ie the tiff and exr are not processed the same by default when you open them…

15

I view both with the same input transform and view transform. In this post it appears it is more about tonemapping.

1 Like
16

I’ve experimented a bit more with this, since I am also seeing a (small) difference between the TIFFs and EXRs when I compare them in python (using pyvips). It is of course always difficult to know if the difference happens during export, or if the library is doing some processing during import…

Anyway, I found the following: EXR export is done using the plug-in mechanism in ART, the code for that is in a separate repo in GitHub - artpixls/ART-imageio: Image loading and saving plugins for the ART raw processor

This looks really nice, since the details of the export can be modified without having to compile the whole application! What I noticed is that the EXR export uses a fixed color profile by applying a small .arp profile before export (ap0-linear.arp). See the documentation in the repo, and the files exr.txt (especially the last line), and exr/exr_io.py (the color primaries for the exported file are defined in the python code in the function write).

The color profile applied by the plugin is RTv4_ACES-AP0. It looks slightly different from the ACEScg-elle-V4-g10 (chromaticity values are slighty different). Perhaps you need to modify the profile ap0-linear.arp to refer to Elle’s profile, and also modify the python code exr_io.py to refer to the same values for chromaticity as the ICC profile uses. I hope this helps.

Continuing this: the chromaticities defined in exr_io.py match values in a table "ACES Primary and White Point Coordinates " in Wikipedia Academy Color Encoding System - Wikipedia. That table has the values for AP1 also, so it should be straightforward to modify the python code to use those values, and the .arp profile to use Elle’s profile. A final step I suppose would be to double check that Elle’s profile uses the same values as the wikipedia table.

As a test I exported from ART using the RTv4_ACES-AP0 to both TIFF and EXR and compared the files in python, and now the difference image is just noise. You could try that to check that there is nothing unexpected going on during import.

Does anyone know or have a source for how the values in ICC profiles translate into the chromaticities stored in an EXR file?

Harri

1 Like
17

Harri, thank you very much for your time digging in this matter. It works really great now !

The last issue is the highlights reconstruction method ! Would you or someone reading this know how to use the same “luminance recovery” highlight’s reconstruction algorithm that is in RawTherapee ? I have tried all the ones available in ART and none match or come close to the RT one.

It is the only difference appearing now :

image
image1712×1142 53.1 KB

From 0.5Gb to 22Mb a file, that is a win nonetheless !

1 Like