On 3/18/2021 4:56 PM, Mike Gordon via olympus wrote:
<<Which makes me wonder if certain lenses plus camera do better or worse with
the
<<AI? Besides just the subject matter. Does that depend on what the AI filter
was
<<trained on?
All speculation but would bet the subject matter trumps almost everything else
as results critically dependent on the training set.
Yup, that's a main reason I chose the subject I did, rather than say a pretty flower or a landscape. Look at the
feathery blue on the dish, then at the hard edges on it. Look at the wood, the drawing on textured paper, the background
to Tom's portrait, the seals. Lots of variation from example to example.
Unlike deconvolution with a known PSF, (where cam/ lens are critical parameters
) the new pixels are totally fabricated based on the surrounding pixels and
training set.
Early versions of Denoise AI had a troublesome tendency to leave noise in dark areas adjacent to bright areas. Seems to
have gone, with updates. Likely fed some additional good examples.
As in all neural networks the exact procedure is a black box.
Yes. That's why I object to the use of the term algorithm applied to this sort of
"AI".
Colorful Box Moose
--
What if the Hokey Pokey *IS* what it's all about?
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|