newbie problem with pixel-shift in RT 5.2


(David Wright) #21

Here’s a close-up using a Sigma DP3M:

David


(David Wright) #22

Try this
https://filebin.net/h0rub1mz011qm3u7


(Andrew) #23

This too seems to have green and magenta bits, and maybe some worm-like artifacts?
But what about the actual subject? - is this the corner of the 4K logo on the cardboard box? If so I’d be guessing it’s a uniform dye colour. Maybe you don’t have it any longer!
Edit - I suppose the worms might be the genuine surface texture?


(Ingo Weyrich) #24

For pixel shift files without motion try this:

psnm

This should give you a good starting point


(Ingo Weyrich) #25

May I ask about the lens used for the Sigma shot? And is the dpi the same as for the K70 shot?

Ingi


(Ingo Weyrich) #26

Your example clearly shows how sensitive rt’s motion detection is :slight_smile:


(Pat David) #27

You should be able to upload both .pp3 files directly here (just drag and drop into the post editor) as well as raw files.


(Morgan Hardwood) #28

Regarding color artifacts, keep in mind that most commercial software does the following behind your back:

  • reduces chromatic aberration
  • reduces fringing (purple and other)
  • denoising
  • probably false color suppression at the demosaicing stage

I’m pointing this out so that you keep in mind that the color artifacts in RT might not necessarily be the result of the pixelshift process. RawTherapee’s PixelShift profiles don’t do these things by default, so it might be incorrect to compare with these filters enabled in one program but disabled in the other.


(David Wright) #29

The DP3M has a fixed 50mm lens. I didn’t make any effort to make the images comparable in size - just wanted to show what a non-Bayer sensor showed for that dot pattern.


(David Wright) #30

I would have to think that the dye is uniform black. The Sigma is also known for green/magenta blotches in the shadows. But I’m thinking that the texture is real. Here’s a blow-up of some text adjacent to this dot pattern:
Sigma


Pentax thru DCU 5.6

Pentax thru RT 5.2


(Ingo Weyrich) #31

It does absolutely not make sense to compare images shot at different reproduction ratios. In your K70 image you can see that each of the dots is represented by ~3x3 pixels. If your sigma shot uses a different reproduction rate (e.g. 5x5 pixels, didn’t count, but seems more than 3x3 at least) for the same area your K70 uses 3x3 pixels you are comparing apples and pears.


(David Wright) #32

Yes, there is that. My interest in this is due to my thought that the result of a pixel-shift Bayer sensor should be just as sharp as the image from a Foveon sensor. But in testing Pentax images processed through DCU 5.2 against Sigma DP3M images processed in the Sigma software, the Sigma images are sharper. Sigma is doing something extra with DPxM images to increase the micro-constrast. My hope was that, with all the other processing features provided in RT, I could reproduce the ‘sigma effect’.

David


(David Wright) #33

I only showed the Sigma image to give an idea what the actual text (or dots) look like. When I actually the two systems I will attempt to have an apples-to-apples comparison. (Acually I’ve done that before, but only using DCU for the Pentax image. I’m hoping that RT will give a better result when I learn how to use it.)

Right now, it seems to me that the RT algorithm for deciding when motion was too much to allow the PS pixel to be used is more sensitive than the DCU algorithm. Otherwise why would DCU show that entire roof using the PS pixels, while RT used a single-shot pixel more often?


(David Wright) #34

Here’s the test I did, almost exactly a year ago. All images at 50 mm; left Pentax pixel-shift processed in DCU 5.2, center Sigma DP3M, right Sigma SD1M:

David


(Ingo Weyrich) #35

Yes, RT at default settings is more sensitive to motion than DCU. In fact, for a lot of subjects, DCU is not sensitive enough to motion and has more motion artifacts. Concerning motion in RT you have several options to increase or decrease the motion detection in custom mode. I suggest to enable the motion mask when you change settings. This way you will see the effect of the settings more clearly. Some of them also have tool tips. I admit that a real documentation is still missing. Didn’t find the time yet to write it :frowning:

Ingo


(David Wright) #36

Thank you very much! This is exactly the information I wanted. Now when I compare the DCU result (left) with the RT result (center) or the no pixel-shift (right), I see that RT gives a result with greater detail w/o sharpening halos. (Maybe too much, but a starting point for further work.) I will examine the settings in that profile. (Is there a way to see a summary of the settings used?)

Thanks again,

David


(Ingo Weyrich) #37

Sure, locate the PS No Motion.pp3 file and open it with a text editor.

Or just look at this:

[Version]
AppVersion=5.0-r1-gtk3-503-g9e1fc78
Version=326

[Sharpening]
Enabled=true
Method=rld
DeconvRadius=0.75
DeconvAmount=75
DeconvDamping=0
DeconvIterations=30

[SharpenMicro]
Enabled=true
Matrix=false
Strength=20
Uniformity=50

[RAW]
CA=true

[RAW Bayer]
Method=pixelshift
PixelShiftMotion=0
PixelShiftMotionCorrection=5
PixelShiftMotionCorrectionMethod=0

(David Wright) #38

Yes, I’ve noticed that the DCU Motion Correction choice often does not find all the motion artifacts, and I am pleased that RT allows adjustment of the sensitivity.

Thanks for that information, and about how to read the pp3 file.

David


(David Oliver) #39

Hi @davidwright. Welcome to the forum from a fellow Pentax user. :slight_smile:

That looks like a great result from RawTherapee. I haven’t used pixel shift all that much apart from some testing yet, but it’s great to see another example.


(David Wright) #40

Thanks.

Pixel Shift and Astrotracer were why I got the Pentax. I currently have only one lens for it, the Sigma 17-70 mm Contemporary.

Here’s some Fall ground cover (no motion). First a thumb-nail image:

Expanded views, with non-pixel-shift on left, ps thru DCU 5.6 in center, and RT 5.2 pixel shift on right. I used the profile Ingo supplied for the RT image, and to the DCU images I applied 0.4 px of Gaussian sharpening. This first comparison is at 200%:

And this is at 400%

If I’m using a tripod, I’m pretty much always shooting pixel-shift. (unless it’s Astrotracer mode…)

David