I finally tried this myself on a closeup of a wristwatch with a moving second hand. The “shadow” from the motion correction where a little disturbing since the second hand had such clear boundaries against a smooth background. Also the exposure where pushed quite a bit which exaggerated the noise somewhat and made the motion correction more visible.
But the smooth transition function fixed that beautifully. Very nice, the sharp transition between the smooth PS background and the single picture very just gone. Well done!
Thanks a lot for the test and the feedback.
Next days I will push a new version with some speedups and better motion detection for high iso images (optionally)
An experimental ISO adaption to increase motion detection for higher ISO. It’s currently mapped to the Experimental checkbox. In my tests it works fine with K-1 files up to ISO 6400. I don’t have examples at ISO 12800 or ISO 25600 from K-1 but for files with ISO >= 51200 it has overdetection of motion. For other models it can be different. At ISO 100 the behaviour is exactly the same as with Experimental disabled.
Tried to take som better shots but various things colluded so that these “studio shots” are the best I have. There’s motion but not quite as complex as my other examples.
I can provide more challenging tests at some other time.
@nosle Thank you very much for making this shots! They are really great and also show that the new experimental iso adaption is better than the old one even for K3II high iso files. Though there’s still room for improvements.
One example, left is with experimental checkbox disabled, right with enabled:
Hi,
I made 2 kits of all possible ISOs. 150 is street scene taken with SMC Takumar 150/4. F values are 7.1 for [100,25600] and 11 for [25600,102400]. 31 – water and ducks with FA30: 7.1 at [100,12800] and 11 at [12800,102400]. Get them at Filebin | 8rfehkjqgykh8jwp
VT
Comparison DCU5 vs Rawtherapee Pixelshift. Both succeeed I guess. DCU appears to do some noise reduction despite disabling it, or it’s the demosaicing. iso 3200
@heckflosse If you still need them, I can upload files from the K-1 tomorrow (will have to ask our secretary which have a cable connection). There are wind turbine in front of cloudy sky + some detailed tree (those are perfectly still). FF, 400-800-1600-12800-25600 ISO.
Looks pretty similar. If you flip between the originals, see link below, you can make out the difference and quite clearly see which areas are motion corrected. They look a bit tighter than the Rawtherapee experimental? what do you think?
@nosle Thanks for the files. They are really interesting
Here’s a screenshot where you can see the enhanced absolute difference between the motion corrected pixelshift dcu and the demosaiced dcu ISO 3200 file. Black means no difference.
Just a thought could the tiled edge come from the texture of the fabric rather than demosaic? It’s quite three-dimensional and the moving shadows could potentially have jagged edges considering the angle of light. Haven’t looked to closely at it as I’m on my phone atm. But it would explain why the smooth surface had sharp edge
I pushed a new option to pixelshift gtk2 branch which allows to use lmmse demosaic instead of amaze demosaic for motion parts. That’s especially useful for high iso shots…
Request for “White Frames” Adding Support for New Raw Formats - RawPedia
for K-1, K-70, and K3II to properly fill up camconst.json and sensor data for pixelshift.
Only categories 1,2 … no need for apperture scaling samples (3). If 1 & 2 is too much, just supply 2 (Long exposures with Long Exposure Noise Reduction ON)
At all ISOs (100-125-160-200-250 … xxxx) preferably in DNG format but 1-2 high ISO samples in both DNG and PEF will help clarify some things.
Preferably in pixelshift mode but if it’s too much single frames will also do the job (just a small uncertainty will remain)
are the best I have. There’s motion but not quite as complex as my other examples.
