Originally posted by reh321 If you evaluate the images [...] on the basis of how the image compares to the actual scene - what should be there and what should not be there - the images come out quite well
[...]
The images should not be evaluated on how they were created - that is how 'religious' discussions get started and how DPR went astray when looking at the K-1ii.
+ over nine thousands.
When you're looking to prove something, you'll succeed.
I once again spent some time comparing the DPR's K-1 and K-1 II's ISO 12800 samples. Silkypix, all NR sliders to zero, starting with identical parameters so that the software would interfere as little as possible. No advanced techniques whatsoever.
At pixel peeping level, the K-1 II image does appear very slightly smoother, but:
- the extra noise on the K-1 image gives a false impression of detail
- the scenes are not absolutely identical (the tiny strand you're looking for might've been moved by air currents - or by the cleaning lady, since the studio scene doesn't get more and more dusty as the years passes)
- the lens is not identical (imagine if we were comparing with the initial K-1 II samples!)
IMHO the differences are within the test's error margin. I'm not saying they aren't real, just that... we're making a mountain out of a molehill. And it's a philosophical "don't touch my RAWs!" issue rather than a practical one.
Wait, it gets better: I nudged the sharpening up a notch in the K-1 II image; I can do that, and noise still remains less objectionable than on the K-1 image.
Or I could apply some (more) NR on the K-1 image to bring it to the K-1 II's level.