Issue with color space changing (ACES AP0 -> sRGB) when converting 32-bit to 16-bit.

[zelenook5F94]

Hello everyone.

I'm wondering if it's somehow possible to achieve what I have in mind:

1. Convert the image to 32-bit color depth.

2. Change the color space to ACES AP0 (I'm using the color profile from the RawTherapee program).

3. And then, convert the image to 16-bit color depth while maintaining the ACES AP0 space, without Photoshop internally reconverting it to sRGB?

What I'm finding is that when I create an action, after the conversion to 16-bit, Photoshop automatically converts the image to the sRGB color space, which has a much narrower gamut than ACES AP0, and that's precisely what I want to avoid. I want to achieve the highest possible quality conversion into a wider gamut to have more latitude for color correction.

I've noticed that with this profile (ACES AP0), Photoshop performs the best color correction using automatic curves, especially for old scanned images. Also, the Kodak Digital ROC plugin (which I use in a separately installed 32-bit version of PS CC2018) responds best to the ACES AP0 space, but unfortunately, it only works up to 16-bit depth.

Please see the screenshot which roughly illustrates what I want, and also shows how PS seems to be forcing its working space. I assume that this procedure would yield the highest possible quality conversion.

If I'm mistaken, and if you are sure that this is absolutely 100% identical to converting to 16-bit depth first and then converting to ACES AP0, please correct me.

P.S. I don't want to tone map HDR because I'm working with SDR photographs.

I am also attaching the color profiles from RawTherapee. ( https://workupload.com/file/w4SUfMNruYt )

Thanks a lot in advance.

---

[davescm · Accepted solution]

When you convert from 32 bits/channel to 16 bits/channel Photoshop will convert the image your current RGB working space, which in your case is obviously set as sRGB.

I've just taken a look at the profile you attached which is not a linear profile, normally used for 32 bit floating point images. It has a transfer curve with gamma 2.4. Such gamma curves are normally used in 8 bit and 16 bit integer images. I would not convert to that gamma profile while working in 32 bit linear.

You could make the RT profile your RGB working space and convert from 32 bit to 16 bit direct into that space. I am puzzled though why you are making the round trip to 32 bit linear at all if you are not using the gains of extended dynamic range that the 32 bit floating point linear space gives. Why not just convert your 16 bit image direct to the RT profile?

Dave

[zelenook5F94]

Most of my photos that I edit are, unfortunately, images from older cameras or compressed JPEGs.
I was thinking hypothetically that by initially converting to 32-bit color depth, I would get the best possible conversion and starting point when converting to the ACES AP0 profile.
But, I tried saving the file converted to 32-bit ACES AP0, and then launching Photoshop with ACES AP0 as the default working space; doing it this way, I obtained the desired 16-bit files in the original color space.
I got absolutely identical edits, both visually and according to the histogram, using Kodak plugins – absolutely identical. So, it seems it's better that I saved myself the trouble and just converted to 16-bit depth and the desired profile in one step.
I was hypothetically wondering if I would gain anything from the extra processing.
At least I tried, thanks for the response. 🙂

[D Fosse]

With a color space as big as ACES AP0 - half of it is imaginary colors outside the visible spectrum - I can imagine that 32 bit depth could be a sensible safeguard against quantization.

With that said, I think all of this is massive overkill. If these are color transparencies, it will be well contained within Adobe RGB. If color negatives, you might need ProPhoto just for processing headroom (not the finished result).

In reality, film technology is severely limited compared to modern digital sensors, both in terms of color gamut and dynamic range. In other words, any current workflow for digital cameras will handle film with no problems (tone curves are quite different and its own separate problem, but you don't need a bigger color space because of that).

Don't know if you're scanning or photographing, but the latter is often much easier to work with if you have a good repro setup.

[davescm]

Given your starting point of older camera images and compressed JPEGs you might want to consider editing in a slightly smaller but still relatively large color space such as ProPhoto, which was developed by Kodak. That is well supported across image editing applications and does not have the potential trip ups of negative primaries that the AP0 space has. Also part of the AP0 space is imaginary, less so with ProPhoto, meaning that your 16 bits are actually used on visible colours in the image rather than being wasted covering the invisible areas.

Edit - Sorry Dag, you replied while I was typing 🙂

Dave

[D Fosse]

Yes, we seem to be saying the same thing. While you could pursue this as a theoretical exercise, in reality we are on a whole other and much lower level of precision. The added precision is an illusion:

https://en.wikipedia.org/wiki/False_precision