JPEGs in AdobeRGB too little saturated

betazoid · January 13, 2020, 6:27pm

https://drive.google.com/open?id=17mF8nalTnzf-MX_8MCcWvMRHsRhlAqkK

betazoid · January 13, 2020, 6:30pm

I can do it though probably to today, but if there is no reaction to a bug report this does not mean much, it certainly does not mean that you should not report bugs any more. Besides, that was right before 3.0 was released and the devs had probably more important bugs to fix. Its clear that they cannot do everything at once.

danny · January 13, 2020, 6:39pm

It may be issue not directly in dt, but in cms library too.

anon41087856 · January 14, 2020, 12:20am

See

github.com/darktable-org/darktable

Internal D65 colour profiles possibly wrong

opened 12:18AM - 14 Jan 20 UTC

closed 07:27AM - 18 Apr 20 UTC

aurelienpierre

no-issue-activity priority: high scope: color management

**To reproduce:** 1. export a a raw picture to TIFF 32 bits float with Adobe RG…B output profile 2. open the exported TIFF (input profile set to "embedded ICC matrix", no module except input/output profiles) 3. export it again with the same parameters. Then, either overlap both exports using the darkroom snapshots or compute the RMSE using GraphicMagick `gm compare -metric RMSE export-1.tif export-2.tif -file diff.png` **Issue:** The blue channel has twice the RMSE as the other channels. On the picture, especially in dark greens, the second export (export from export) displays darker colours. With LCMS2 on (CPU SSE2 path): ``` Image Difference (RootMeanSquaredError): Normalized Absolute ============ ========== Red: 0.0000082394 0.5 Green: 0.0000052063 0.3 Blue: 0.0000157842 1.0 Total: 0.0000107103 0.7 ``` With LCMS2 off (OpenCL path): ``` Image Difference (RootMeanSquaredError): Normalized Absolute ============ ========== Red: 0.0000067537 0.4 Green: 0.0000059158 0.4 Blue: 0.0000203831 1.3 Total: 0.0000128592 0.8 ``` sRGB exhibits the same issue: ``` Image Difference (RootMeanSquaredError): Normalized Absolute ============ ========== Red: 0.0000119955 0.8 Green: 0.0000108273 0.7 Blue: 0.0000247451 1.6 Total: 0.0000170630 1.1 ``` Comparing with an external ProPhoto ICC profile: ``` Image Difference (RootMeanSquaredError): Normalized Absolute ============ ========== Red: 0.0000011652 0.1 Green: 0.0000012465 0.1 Blue: 0.0000031726 0.2 Total: 0.0000020798 0.1 ``` Diff image (differences highlighted in red): ![diff](https://user-images.githubusercontent.com/2779157/72302764-5c876880-366b-11ea-9ce6-46e3c0e4081b.jpg) I suspect an error in the internal Adobe RGB and sRGB profiles on the B channel coeff. Based on @blitzgneisserin https://discuss.pixls.us/t/jpegs-in-adobergb-too-little-saturated/15785 (but not sure it's the same problem).

danny · April 11, 2020, 5:13pm

I opened the issue on github.
It seems that the input color profile module doesn’t apply the embedded profile by default.
https://github.com/darktable-org/darktable/issues/4651

betazoid · April 11, 2020, 5:20pm

I thought this was fixed weeks or even months ago.

danny · April 11, 2020, 5:23pm

It’s still in current master and also in released 3.0.1

hannoschwalm · April 18, 2020, 7:13am

It has now been fixed in dt master.

danny · April 18, 2020, 11:49am

Thank you