Support for Pentax pixel shift files #3489

(Andrew) #21

Wow, I’m impressed how you understand the processing needed and just got this done. It’s such an improvement. Well done.


I test my DNG in RT but DCU make jpg better with small movements :wink:
Im main window RT I see small red frame in this DNG with movements ?
I disable auto CA correction, hot pixel correction.

(Ingo Weyrich) #23

As I wrote above: Current implementation does not support any kind of movement detection/correction.
For the small red frame please show a screenshot.


This frame be red, green or black ? :wink:

It is visible at zoom ?

Try this DNG self :


(Ingo Weyrich) #25

I tried with your file. No frame visible here. Can you post the section of your camconst.json for K-70 please?


After restart I don’t see this frame :wink:
Green frame is visible after fast zoom mouse wheel…see it 0,25 sek. and hide after image zoom process end.
Try You add automatic detection pixel shift DNG and automatic select correct good demosaic metod…
Maunal switch is slow when I have many varius DNG in one folder…

(David Oliver) #27

IMGP0569.DNG (Pentax K-3 II, SMC Pentax-DA 35mm F2.8 Macro Limited, ISO 100, f/8)

There’s what I would describe as a small amount of chromatic abberation around some edges, including the outline of the boots against the wall. I’ll try to get a shot/shots with more CA soon; I also have an old Pentax 50mm prime I can try.

Processed very quickly with only basic adjustments and auto correction features turned off in Pentax/RICOH/Silkypix Digital Camera Utility 5.

Left: PS off; right: PS on. No additional sharpening applied. (Click for full size to see more of a difference.)


(Ingo Weyrich) #28

Thanks for uploading the file. It’s a nice example.

First there is a stuck pixel which clearly shows that I have to address the .badpixels processing which is currently disabled for pixelshift.

Second it shows that even smallest movements of the subject results in artifacts.

Here are the 4 frames of a pixelshift shot showing a moving part.

And here a screenshot showing the artifacts (left is 1st frame of pixelshift demosaiced using amaze, right is the combined image with the artifacts)

(Ingo Weyrich) #29

Can you please rephrase? I don’t understand. Thank you!

(Ingo Weyrich) #30

There is no single correct method for pixel shift files. You can demosaic each of the frames using one of the standard demosaic tools and also can combine using pixel shift algorithm. Each of them is correct. But I’ll ad at least a hint in info dialog to show whether a file is a pixelshift file or not.

(Ingo Weyrich) #31

btw: If you select pixelshift on a non pixelshift file it falls back to amaze demosaic automatically.

(Ingo Weyrich) #32

Is pixelshift overkill?

Here’s a screenshot showing Amaze from 1st frame compared to pixelshift. Which is which? Isn’t pixelshift pixel peeping par excellence most of the time? Of course there are use cases for it as I demonstrated in the examples some posts above, but for everyday shots?

(Jean-Christophe) #33

Pixelshift is of course not for everyday shots. You have to prepare the settings (tripod, set 2s delay, static scene). It’s not overkill, just for artistic scene or pack shot with continuous light.

(Ingo Weyrich) #34

I pushed some changes to the pixelshift branch. Still no movement detection, but faster loading (decoding) of pixelshift files, fixed a bug and averaged the green channels.


TNX Ingo !
You make good work !

(Ingo Weyrich) #36

I tried pixelshift on an ISO 51200 file. Left is amaze demosaic, right is pixelshift.


Compare last RT compilation and DCU with motion correction on/off.

Only AWB and deconvolution sharpen in RT and only AWB in DCU :wink:!3Ql1wKJL!HcFSJFEFavMuT9UMyZTRig

(Ingo Weyrich) #38

Thanks for the files :slight_smile: Can you add an amaze demosaiced version please?

(Ingo Weyrich) #39

Looking at the cat’s nose in 200% view I have the impression that DCU applies some additional processing. Though rt pixelshift still shows more artifacts than DCU, the tip of the nose has better detail in rt version.

(Ingo Weyrich) #40

In the pp3 you attached I saw that you enabled some stuff which makes it really hard to compare the results (like sharpening and auto exposure…) I would suggest to use the RT neutral profile + changed demosaic method for comparisons.
Is it possible to make a neutral processing (without sharpening and so) on in DCU?