Changes in Demosaic>method: photosite color (debug)?

Software darktable
1

In a presentation I am preparing, I want to show the pixel pattern on the sensor. So I load an image into DT, turn off all modules and set demosaic to “photosite color (debug)” and zoom in to 1600%. I first did this about a year ago, so probably in darktable 4.2, and I got this:

Bayer-mønster
Bayer-mønster1524×973 1.87 MB

Now I did it again in 4.6.0 with the same RAW file and I get this:

Sensor-pattern-DT-4_6
Sensor-pattern-DT-4_61920×1080 1.45 MB

This example is from a Canon camera (Bayer pattern), but something similar happens to Fuji files. Now, in my mind the old version is the correct one, with the R, G and B pixels nicely laid out next to each other. In the new one there is something strange going on. The R, G and B pixels seem to be smaller, and with some darker green in between, and this green sometimes turns towards red or blue.

So I just wonder what changed, and why?

I could not find anything in the release notes for 4.4 or 4.6. The manual just says:

photosite_color is not meant to be used for image processing. It takes the raw photosite data and presents it as red, blue or green pixels. This is designed for debugging purposes in order to see the raw data and can assist with analysis of errors produced by the other demosaic algorithms.

I think this aligns best with the oldest example.

2

Strange, if I try what you describe, I still get the pattern in your first image, and that is using version 4.6.0

3

That’s really weird. The only other thin I can think of is a change of OS. The first one was done on my Linux machine and the new one on Mac. I can’t see how that would affect anything, but I’ll try on Linux again when I get home from work.

4

With 4.7.0 and Windows, I get at 100%

image
image1312×890 1.31 MB

and zooming to 800%
image
image1321×900 72.2 KB

There is a strange dot pattern visible.
In the image, there is no such pattern (100% zoom)
image
image1311×890 124 KB

5

Hi, I observed the same pattern (small pixels…) as you on iMac intel OS 14.3 with the nightly builds. You can get back to the other pattern if in preferences/processing you select “always use Little CMS to apply output color profile”. Interestingly, if you then deselect the option, at least on my Mac, you don’t get back the little pixels pattern. I don’t have any explanation, but certainly here are developers that may explain this observation.

6

Do you have lens correction turned on? That will cause that kind of pattern.

7

When I do this, nothing happens. The small pixels pattern stays the same.

And on my Linux box (Linux Mint), also with DT 4.6.0, the pattern is the correct one, with the large pixels. So it seems to be a Mac issue. My MacOS is Sonoma.

This is very strange, I really hope someone is able to clarify, although it is mostly out of curiosity.

8

Just in case that a developer reads the post, after enabling the “littleCMS pref” I restarted dt (as requested) to make it effective, and get the “old pattern” of pixel display. The message “needs a restart” was shown only when I selected littleCMS the first time. Subsequent changes to enable or disable this preference did not show the message.

9

I did not get the “needs restart” message, but I did anyway, just in case. And no change. But I discovered something else: When I have normal demosaicing on and stay zoomed in at 1600%, the image on the Mac has much higher resolution than the image has in reality.

I tested on another image. I exported a .tif of the image, then zoomed in to 1600% on the RAW and took a screenshot. Then I opened the two in Preview and laid them over each other:

Screenshot 2024-01-29 at 19.56.09
Screenshot 2024-01-29 at 19.56.092270×1740 2.45 MB

The small one in the foreground is the screenshot, it is overlapping pretty exactly the .tif. You can see the resolution is about the double.

So there seems to be some interpolation/“resolution enhancement” going on when zooming in on the Mac, and I don’t like it. If I zoom in this much, clearly it is because I want to see the individual pixels, and not something like the “can you enhance that?” from the movies :thinking:

Aaand, I just confirmed the suspicion I have had for a while, that the middle click zoom on the Mac zooms to 50% and 100%, and not to 100% and 200%. The preference “middle mouse button zooms to 200%” is checked.

10

This is what I get with 4.7.0+391~g99e5dae3c1:

image
image1446×1049 24.8 KB

However, lens distortion control (and sharpening, and so on) must be turned off, otherwise the pixels will be warped. With lens correction on:

image
image1446×1049 29.4 KB

11

Yes, I know about the lens correction. I have it off.

12

Now it’s getting late and I am tired and confused. But my conclusion so far is that the Mac is messing with scaling, zoom levels and pixels.

I said above that the Mac middle mouse button zoomed to 50 and 100%. Now it’s back to saying 100% and 200%, but in reality it is 50% and 100%.

This is zoomed to 100% on Linux:
image

And this is “100%” on the Mac:
Screenshot 2024-01-29 at 21.53.29

These are on the same external monitor with the same resolution, and the same darktable layout, just slightly different width of the left panel. Height is the same. Raw file is 6000x4000.

Now I’m off to bed! Good night.

1 Like
13

I didn’t know the impact of the lens correction module. When I disabled it, The image looks correct.

image
image1318×897 51.4 KB

There is still something on the last row and column, the pattern is not correct. But this is another issue.

14

I just installed darktable on another Mac, a Mac Studio at work (M2 Ultra, 64 GB) . Everything works as normal there, so I guess it’s some graphics thing in the MacBook Pro (M1 Max, 32 GB)…