Gain map/PNG gamma issue

[macdavis]

It seems the gamma of the P3 SDR base image produced by Adobe Lightroom is 0.455 (inverse of 2.2). PNG decoded from libjxl preserves this data, however, it looks like libavif does not, (Edit: irrelevant, but might also be something to look into) resulting in gain map gamma issue when combining base and alt PNG images (Gain Map Gamma is always equal to 1).

Way to reproduce

1. Download avif gain map sample file from https://helpx.adobe.com/camera-raw/using/gain-map.html. In this case, I use 01_base_hdr.avif as an example.

2. avifgainmaputil printmetadata 01_base_hdr.avif

Gain Map Gamma: R 0.25 (as fraction: 268435456/1073741824) G 0.25 (as fraction: 268435456/1073741824) B 0.25 (as fraction: 268435456/1073741824)

3. avifgainmaputil extractgainmap -q 100 01_base_hdr.avif 01_base_hdr_original_gain_map.jpeg

4. avifgainmaputil swapbase -y 444 -q 100 --qalpha 100 --qgain-map 100 01_base_hdr.avif 01_base_sdr.avif

5. avifdec.exe -d 16 01_base_hdr.avif 01_base_hdr.png

6. avifdec.exe -d 16 01_base_sdr.avif 01_base_sdr.png

7. avifgainmaputil combine --qgain-map 100 --depth-gain-map 10 --yuv-gain-map 444 -q 100 -y 444 -d 10 --cicp-base 9/16/9 --ignore-profile --cicp-alternate 9/1/9 -s 10 01_base_hdr.png 01_base_sdr.png 01_combined_base_hdr.avif

8. avifgainmaputil printmetadata 01_combined_base_hdr.avif

Gain Map Gamma: R 1 (as fraction: 1/1) G 1 (as fraction: 1/1) B 1 (as fraction: 1/1)

9. avifgainmaputil.exe extractgainmap -q 100 01_combined_base_hdr.avif 01_combined_base_hdr_gain_map.jpg

Both gain maps should be quite similar, but they are not, as shown below.

[maryla-uc]

Hi, thanks for the detailed bug description. Indeed, libavif's current gain map creation function always uses a gamma of 1. However, this shouldn't be a problem in practice (unless you're actually facing issues/display differences). The gainmap's gamma is only used as a compression/precision optimization.

In general, the gain map's metadata heavily affects how the gain map image looks, but it won't necessarily result in a significantly different tone mapped image.
I tried changing the gamma to 0.25 and the resulting gain map still looks fairly different.

What I DO notice is that there seems to be a color shift from the original. The green is less intense. I'm not sure what's causing this and would need to investigate more.
Another thing that can cause differences is the interpretation of the transfer characteristics value "1" (BT709), since not everybody uses the same function for it, and the one in libavif is different from the one in chromium (which inteprets it as sRGB, see https://chromium.googlesource.com/chromium/src/+/lkgr/ui/gfx/color_space.cc#895)

[macdavis]

Thanks! I did notice some subtle differences and that's what made me look into it. Is it possible to make gainMapGamma assignable, similar to CICP?

With regard to the color shift, I think it is indeed related to the interpretation of the transfer characteristics value "1" (BT709). I loaded 01_base_hdr.avif into Adobe Lightroom and exported 3 photos in JPEG, TIFF and AVIF (Transfer Char. = 1) respectively, all in single layer SDR Display P3 Gamut. The color decoded via libavif is very different from the others. I've attached JPEG and AVIF at the end.

Maybe implementing a combine function of "Base Image + GainMap Image + GainMap Metadata" (similar to libultrahdr API - 4) can work as a bypass?

01_base_sdr_Lightroom.zip