Support for Pentax pixel shift files #3489


#81

Best resolution in RT ( 7384x4942 vs DCU 7360 x 4912) and best effect of correction !


(Gimbal Lock) #82

Nice. Seems very promising indeed.


(Ingo Weyrich) #83

7384x4942 was due to an old camconst.json file used for pixelshift. Final resolution will be 7368x4924


(nosle) #84

I pulled and built the latest pixelshift branch yesterday and the motioncorrection preview looked great for my Pentax K3 II files if careful to switch back and forth between images to get rid of the colour shifts across the whole image.

When exporting jpg’s however I end up with a pink version of the green motion preview pattern. This is not visible in the preview.

I fully understand I’m messing with software in motion but I see you managed to get images out of the program. Is the pink patterns something you see as well or a bug I should investigate?


(Ingo Weyrich) #85

@nosle

SaveAs has a bug for pixelshift which will be fixed in final version. As a workaround you can save using the queue (Add to queue, start queue)

Edit: Thanks for testing :slight_smile: If you can provide pixelshift files for our tests, that would be very welcome too :sunny:


(nosle) #86

Where do you want your pixel shift files delivered?


(Ingo Weyrich) #87

filebin.net if possible


(nosle) #88

4 Pentax K3 II pixel shift files at https://filebin.net/zy5q2lsbbbruv1ti
Mostly foliage movement but also a van speeding past, people and cranes moving in the background but I guess the “Show motion” option will make that instantly obvious :wink:


(Ingo Weyrich) #89

Great, thank you, already downloading :slight_smile:


(Ingo Weyrich) #90

That’s a great test file!


(Ilias Giarimis) #91

Τhanks also …

If you can provide a set of pixelshift Black Frames at all ISOs (integer and intermediate) will help for better adaptation …
Just go in a dark room, put the lens cover on and shot with fast shutter speed (1/500 or more) and closed aperture (F16.0 ?).
Compress the samples with 7z (super compression, PPMd method for best results) before uploading to save uploading time.

Next will come a request for White Frames at all ISOs (normal shots no pixelshift …) :wink:


(nosle) #92

25 Pentax K3II pixelshift black frames as requested

https://filebin.net/hyxw4ieb1wczre6g

All shots taken at 1/500 f/22. Didn’t use a viewfinder cap so if the process is extremely sensitive some light might theoretically leak in through the viewfinder. There was a tiny leak of light into the room from above the closed door. However at 1/500 it shouldn’t be an issue?


(Ingo Weyrich) #93

I just pushed a fix which gives better result for the van. Left is without motion correction, middle is version from yesterday. right is new version


(nosle) #94

Looking good! When trying to get rid of the last bits of artifacting yesterday I noticed that the “False color” slider did the job. I was not expecting that result but I can’t say I understand what’s going on…


(nosle) #95

I found another file with a difficult test case. This time its soft shadows moving on a facade. There’s a “hole” with rather sharp edges left over after the motion correction. Very different case from speeding van.

https://filebin.net/pcjvbwsmz44ifeik


(Ingo Weyrich) #96

Hmm, can’t reproduce that here. Which revision was that and can you post the pp3 please?


(nosle) #97

https://filebin.net/wkqix2e3izrl6adc
It’s not the identical pp3 as the above screenshot but the problem is still there. Could be any other parameter causing the issue of course. The top commit is 28358d6bf2d3d6826ef4b03ec71537f4ecae3054 pixelshift: added V4 for tests.

Something I’m doing is causing trouble see this image as well. This kind of stuff wasn’t there before…


(Ingo Weyrich) #98

Thanks, I’ll have a look.


(Ingo Weyrich) #99

@nosle:
I had a look using current pixelshift branch but could not reproduce the issue, strange… but
I saw you used retinex and sharpening too, better to disable this tools for comparisons of pixelshift images. Best is ‘neutral profile’ + pixelshift


(Gimbal Lock) #100

Without knowing any details, could it be that when using image data from one raw due to motion you don’t necessarily get the exact same shade of blue (in this case) as the surrounding blue made up of the full pixelshifted data?