Sometimes 1-Pixel-Artifacts from X-Trans Demosaicing

Software RawTherapee
21

I hope this is the right thread, I’m not sure if this is the same issue or not. I compiled the lastest dev version and I got this with my X-T30. The Markensteikn algorithm produces strong colour artifacts in high contrast areas in some directions. This was already the case with a version compiled one or two weeks ago. Similar artifacts are visible with all other algorithms. Maybe it’s a problem with the colour pattern itself?

markesteijn
markesteijn1536×833 104 KB

This is less important of course, but the “none” method also doesn’t show the right colour pattern:

none
none1537×836 106 KB

22

This looks like CA to me.
Are there less “artifacts” in the center of the frame?

23

Wouldn’t CA show up everywhere on the roof, not just in the right slope? And wouldn’t the colours be the same on both edges? Also, this was taken at f/7.1.

edit: wait no you’re right, the colours wouldn’t be the same. But still, I’m not convinced it’s CA.

24

Using ‘none’ you should also use neutral profile to see the xtrans patterns (so neutral + demosaic none)

25

Well, I can show you files from my Nikon D700 (bayer) which have a large amount of CA even at f/8.0

26

Okay I found the neutral profile and that works! I need to read the manual more, I don’t understand what is going on here.

I will do more tests to see if it can be CA. Unfortunately I don’t have other lenses.

27

I agree, that the change here looks somehow wrong

grafik
grafik1536×833 96.4 KB

28

Exactly! I thought it had to do with alignment with the sensor matrix, but maybe there is another explanation.

29

Can you please show where in the frame this part of the image is?
Is your lens the 35/F2?

30

It’s the f/1.4.

It’s circled here:

roof
roof900×607 686 KB

1 Like
31

The 1.4 has quite low CA. This looks strange

32

Actually it might be transverse CA! I didn’t expect it to be so bad. I have checked other pictures and It looks like the blue is always moved towards the centre of the picture.

33

I rarely use lensfun or explicit CA correction with this lens. It is simply not necessary there.
Could you post a RAF?

34

Maybe it’s just a bad copy? CA correction in RT makes it much better even though it’s not completely gone. DSCF0468.raf (28.5 MB)

35

I just got this lens as well, and I have an X-T20 if you guys need another test point.

1 Like
36

I think you should not use such extreme magnification.
Here is an example of my copy mounted on a X-E2s, also using 1600. The snippet is from far right of the frame:

Schlat
Schlat1522×996 321 KB

And this is the same with 100%

Schlat100
Schlat1001532×991 499 KB

There is visible CA, which is being over-corrected by lensfun.

Iridient X-Transformer performs a bit better:

Schlat_iri
Schlat_iri1531×992 272 KB

37

Oh I have to apologize @fschhh! The blue glow on your image does not really look like CA. At least not what I am used to. It is stronger and shows a very saturated blue, my copy shows a bit blue towards the center and stronger orange outwards.

Edit: playing with Defringe tool

  1. disabled

    blue1
    blue1930×621 129 KB

  2. enabled

    blue2
    blue21031×766 168 KB

The GUI for Defringe seems to have a bug. It shows only the upper half of the equalizer for a few seconds:
defringe

1 Like
38

I can confirm this bug, please file a bug report. (I apologize for not reporting it sooner, I’ve been lazy)
The same bug also appears on the skin tones graph of the vibrance tool.

39

Please do not report this GUI bug. It is a well known bug for all curve dialogs, for which we haven’t got a solution as far as I know :slight_smile:

1 Like
40

I checked, I really think it’s CA now. The bright areas are always leaking blue towards the centre of the image. My copy of that lens must just be worse than yours, and I’m not used to such high resolution sensors (my previous camera had 12 Mpix). Sorry for polluting the thread.