You’re right that the discrepancy is due to the jpeg compression, but human color vision is a super complex thing, and the memetic function of a photographic image production system also implicitly compresses the range of brightnesses in the scene to something that “looks right” for screen or paper. There’s nothing “objective” about photography.
It’s not about JPEG compression, it’s about highlight clipping.
I assure you that if you held these photos up next to the actual scene, the tonemapped darktable version would reflect the real world color better, simply because it retains the correct hue, which is roughly the same as the darker surroundings, instead of hue shifting to yellow.
It’s me again. I have developed yet another module for darktable. Here is the basic result:
Stay tuned to see if it will get merged like the 2 previous.
Very impressive, particularly with the shadow recovery. The accuracy of the colours in the recovered shadows is spot on!
What is this latest module you’re working on?
I think the dynamic range compression in Rawtherapee could works in a similar way of the log+curves workflow.
It’s necessary to add only exposure compensation before the dynamic range compression(like the middle gray slider in the darktable log module).
In this image I have increased the exposure in darktable about 2.22 EV and exported as linear floating point, then in RT the dynamic range compression tool with strenght=35, details=75 and anchor=9
Add a single s-curve