search

Beep6581 / RawTherapee

A powerful cross-platform raw photo processing program
https://rawtherapee.com
GNU General Public License v3.0
2.87k stars 322 forks source link

constrast by detail levels reduces saturation / fit-to-screen preview wrong #6327

Open blitzgneisserin opened 3 years ago

blitzgneisserin commented 3 years ago

Short description If contrast by detail levels is active, saturation of the image is reduced. The fit to screen preview shows an image that is too saturated (apparently it shows a preview without applying contrast by detail levels, even if it is active, or color management is off). The exported image has the same saturation that the image has when it is viewed at 100%-view. However, it is difficult to judge the saturation of the whole image at 100% view.

Steps to reproduce

  1. open a photo (raw)
  2. apply contrast by detail levels, e.g. level 0 value 4, level 1 value 3, level 2 value 2, level 3 value 1 (etc)
  3. zoom in to 100% view
  4. zoom out to fit to screen view
  5. compare the colors
  6. export the photo
  7. compare the colors of the exported photo with the fit to screen preview and the 100%-view
  8. switch off contrast by detail levels
  9. export the photo
  10. compare the colors of the two exported photos

Expected behavior The photo should have the same saturation when viewed at 100% view, fit to screen preview and exported.

Additional information I have tested this on a profiled BenQ SW240 screen which has a gamut that is slightly larger than AdobeRGB. Not sure this behavior can be observed on sRGB screens. Using RT dev Appimage from June 14, 2021, on Linux Mint 20.2, graphics card Nvidia GTX 1660 Super, AMD Ryzen 7 processor, 16 GB of RAM

Other useful information: Unfortunately I cannot document this properly because for some reason the screenshot that I take looks less saturated in an image viewer than what I actually seen in RT. The screenshot actually has the same saturation as the exported photo. This is the exported photo when cbdl is on: P6232299-8 This is the exported photo when cbdl is off: P6232299-9 This is the raw file + the 2 pp3 files: rawpp3.zip

Thanatomanic commented 3 years ago

@blitzgneisserin Can you confirm that the exported image looks the same as the 100% zoomed preview in RawTherapee? That should be the case at least. Please be aware that due to the nature of the CBDL algorithm, the zoomed-out preview cannot be fully accurate with respect to the exported image. That's why there is a 1:1 icon on the tool.

blitzgneisserin commented 3 years ago

@Thanatomanic I can confirm that the 100%-view is correct and identical with the exported image. I am aware that the zoomed out preview is not entirely correct when CBDL is active, but it should not affect saturation, hue or brightness when the whole image is vieible. I think this is a bit annoying. Please try to think about something.

Lawrence37 commented 3 years ago

Starting from the neutral profile and applying only CBDL, the chroma does not change. Exporting the image with the supplied pp3 without CBDL then applying CBDL on the exported image also produces the expected result. Disabling Exposure > Contrast in the pp3 results in no chroma change.

My conclusion is that CBDL does not affect chroma. Contrast does change chroma. If CBDL comes before contrast, then changing CBDL will affect the input for contrast and therefore result in a chroma change. The only problem with my theory is that CBDL supposedly goes after contrast.

blitzgneisserin commented 3 years ago

Starting from the neutral profile and applying only CBDL, the chroma does not change. Exporting the image with the supplied pp3 without CBDL then applying CBDL on the exported image also produces the expected result. Disabling Exposure > Contrast in the pp3 results in no chroma change.

My conclusion is that CBDL does not affect chroma. Contrast does change chroma. If CBDL comes before contrast, then changing CBDL will affect the input for contrast and therefore result in a chroma change. The only problem with my theory is that CBDL supposedly goes after contrast.

Thanks for having a closer look! I am not quite sure about this: it seems that avoiding all rgb contrast and saturation silders (i.e. except contrast in lab and the color and light module in local adjustments) prduces a more acceptable result. But I think a slight reduction of saturation is still there.

Lawrence37 commented 3 years ago

Can you share a pp3? I have found no differences in chroma (aside from rounding errors) after disabling contrast in the original pp3. I verified by exporting 16 bit tiffs and extracting the chroma component in GIMP. Comparing by eye is no good because the visual system perceives less saturation when there is an increase in contrast (e.g. using CBDL or Lab L curve).

blitzgneisserin commented 3 years ago

Can you share a pp3? I have found no differences in chroma (aside from rounding errors) after disabling contrast in the original pp3. I verified by exporting 16 bit tiffs and extracting the chroma component in GIMP. Comparing by eye is no good because the visual system perceives less saturation when there is an increase in contrast (e.g. using CBDL or Lab L curve).

Well, I can definitely confirm that there is no difference in chroma when contrast and/or saturation are disabled. I will share a pp3 later today. I need to go to the swimming pool now before all the other people arrive. But I want to add this: what matters to me and I think most other users is not how much saturation/chroma is really there but how much is visible.

Thanatomanic commented 3 years ago

I did not do any visual testing myself, but

the visual system perceives less saturation when there is an increase in contrast

and

what matters (...) is not how much saturation/chroma is really there but how much is visible.

don't seem reconcilable? Adding a saturation change to counter the apparent change due to the algorithm seems a little contrived.

Lawrence37 commented 3 years ago

But I want to add this: what matters to me and I think most other users is not how much saturation/chroma is really there but how much is visible.

For this, you may want to try wavelet levels contrast and chroma (Advanced tab > Wavelet levels > Contrast; Advanced tab > Wavelet levels > Chroma). Set "Apply to" to "Whole luminance range" and increase "Chroma-contrast link strength".

blitzgneisserin commented 3 years ago

ok, well I am glad there are any reactions at all.

I think just adding saturation would not be optimal. I think the reduction of chroma is stronger in regions where there is more contrast/details, such as... trees and bushes e.g.. I think the module was "ported" from darktable's contrast equalizer. There the fit to screen preview is correct and it is visible that chroma is reduced. So the user can increase chroma if necessary. But there are issues with the preview there, too, though not with chroma. I think hanatos wrote somewhere that the preview is "not fun to code". There is a similar module in art, but there, the preview is totally crazy. I think there, chroma is correct but sharpness and contrast not at all.

But it is an important observation that chroma is not reduced if modules that change contrast and/or chroma are not active. So either this can be fixed with some change in the position of this module (which modules are before and after it), or it is some kind of "calculation error".