Strange color artifacts in Jupiter

[luisfl]

RT version: rawtherapee-5.4-5.7.x86_64 in fully updated Fedora 28 (distro RPM package).

When I open this ORF image of Jupiter with RawTherapee,

http://luis.impa.br/photo/lixo/Dated_JN-180601-B_21202.orf

RT shows lots of color artifacts, independently of the demosaicing algorithm.
The same RAW file processed with Darktable or Dcraw (through UFRAW) shows no artifacts (see attached jpg crops at 800%).
I have only observed all these artifacts with astrophotography of planets, and only recently. It seems old RT versions did not show this problem.

[agriggio]

are you using auto-ca correction by any chance? (just guessing)

[Thanatomanic]

That doesn't seem to matter. I tested this on the latest build and I can reproduce the weird behaviour with all demosaicing modes and with everything else off...

[gaaned92]

VNG4 seems OK

[Beep6581]

darktable uses PPG by default for .orf files.

[Floessie]

Color aside: At least you can spot Ganymed on the RT rendition. 😁

[agriggio]

Try with green equilibration set to 100% and a couple of steps of false colour suppression:

• AMaZe:
  

• LMMSE:
  

[Beep6581]

@luisfl could you upload some other shots from this camera - both astro and non-astro?

[heckflosse]

Neutral profile + following patch gives this

diff --git a/rtengine/green_equil_RT.cc b/rtengine/green_equil_RT.cc
index 934b7df4a..9706b06c1 100644
--- a/rtengine/green_equil_RT.cc
+++ b/rtengine/green_equil_RT.cc
@@ -46,13 +46,14 @@ void RawImageSource::green_equilibrate_global(array2D<float> &rawData)
 
     for (int i = border; i < H - border; i++) {
         double avgg = 0.;
-
+        int ng = 0;
         for (int j = border + ((FC(i, border) & 1) ^ 1); j < W - border; j += 2) {
-            avgg += rawData[i][j];
+            if(rawData[i][j] > 0.f) {
+                avgg += rawData[i][j];
+                ng++;
+            }
         }
 
-        int ng = (W - 2 * border + (FC(i, border) & 1)) / 2;
-
         if (i & 1) {
             avgg2 += avgg;
             ng2 += ng;

[agriggio]

@heckflosse seems good to me! 👍

[agriggio]

Though I'm wondering whether this new sensor needs the global green equilibration at all...

[luisfl]

Sorry for the delay. Answering the comments (some users appear twice, please read until the end):

@Thanatomanic : Precisely. Everything off.

@gaaned92 : VNG4 is quite bad. It unacceptably blurs fine detail.

@Beep6581 (first comment): Amaze in Darktable gives even better results, showing two bands of clouds in jupiter(!) (and no color artifacts of any kind, everything turned off).

@agriggio (first comment): Sorry, but I think yours is the worst possible solution. I have seen many times RT false color suppression introducing lots of false color artifacts. I don't even look anymore to that option. About green equilibration, it seems to me it just blurs the artifacts. I can see horrible color halos.

@Beep6581 (second comment): Sure (I'm posting the moon shot since I see lots of color artifacts lately with RT, even at lowest ISO setting):
http://luis.impa.br/photo/lixo/teque-teque_SE-180330-B_17480.orf
http://luis.impa.br/photo/lixo/lua_JN-180424-B_19844.orf

@heckflosse : I still see lots of color artifacts, although certainly much better. Darktable and DCRAW still show a much cleaner image. I am not sure that applying green equilibration (as in your patch) is the solution to this bug...? You could have also applied defringle. I'm not sure the bug should be closed.

@agriggio (second comment): I have never used it. This bug was introduced in the last version(s?) of RT.

[luisfl]

Please, let me show an "everything off" version processed with darktable (AMAZE); observe the two vertical bands of clouds in jupiter, more noticeable than in the RT version: