Chromatic aberrations issue?

gadolf · June 21, 2020, 4:17pm

When I turn chromatic aberrations on, there’s a noticeable decrease in saturation of the reds and greens/yellows on bright areas.
It seems to me as if the module doesn’t apply the effect only to the high contrast borders, where we expect to find the chromatic aberrations, but spreads into flat regions of the image.
Is it by design?
this is darktable 3.1.0~git2163.fda1e5796

CA on

CA off

afre · June 21, 2020, 4:40pm

Read https://www.darktable.org/usermanual/en/correction_group.html#chromatic_aberrations first. May be a preview problem; or perhaps, the detector needs an overhaul.

gadolf · June 21, 2020, 7:16pm

Thanks @afre.
I already removed the zoom images from the original post and exported the two versions.
But the differences between the exported jpegs remain as show in the zoomed out images left in my post.

afre · June 21, 2020, 7:23pm

Could you post the exports? 1:1 crops are fine.

Also, a sample file and sidecar, for those who want to investigate?

gadolf · June 21, 2020, 7:28pm

Here they are:

CA on:

CA off:

IMG_7739.cr2 (11.4 MB)
IMG_7739.cr2.xmp (11.0 KB)

This file is licensed Creative Commons, By-Attribution, Share-Alike.

Jade_NL · June 22, 2020, 7:14am

Using your RAW I can reproduce this, with or without your accompanying xmp file. You can also reproduce it when only applying the bare essential default modules + the ca module. So it does not seem to be a module combo thing.

~~However: I cannot reproduce this with any other RAW file. Tried using cr2’s, nef’s and orf’s and I do not see this phenomena.~~ Actually I can see this on other RAW’s, but not on all of them.

Tested using dt 3.1.0+2350~g3f399bdf8

EDIT: This isn’t 3.1.0 specific. I just tried 3.0.2 and it also shows up in that version.

heckflosse · June 22, 2020, 10:04am

If this is about raw ca correction, then may have a look here. RT had the same issue

gadolf · June 22, 2020, 10:55am

Thanks @Jade_NL and @heckflosse, I opened an issue: CA corrections desaturate image · Issue #5604 · darktable-org/darktable · GitHub

heckflosse · June 22, 2020, 11:35am

@gadolf Here’s a poiner to the part of the RT code which tries to avoid the desaturation

github.com

Beep6581/RawTherapee/blob/dev/rtengine/CA_correct_RT.cc#L1274

    
      
          #endif
                              for (; col < width - cb; col += 2, indx++) {
                                  rawData[row][col] = std::max(0.f, RawDataTmp[indx]);
                              }
                          }
          
          
            }
                      // clean up
                      free(bufferThr);
                  }
                  if (avoidColourshift) {
                      // to avoid or at least reduce the colour shift caused by raw ca correction we compute the per pixel difference factors
                      // of red and blue channel and apply a gaussian blur to them.
                      // Then we apply the resulting factors per pixel on the result of raw ca correction
          
          
#ifdef _OPENMP
                      #pragma omp parallel
          #endif
                      {
          #ifdef __SSE2__
                          const vfloat onev = F2V(1.f);