On 8/26/2018 3:18 PM, Moose wrote:
On 8/26/2018 3:01 PM, Tina Manley wrote:
<<<I'm trying out the new Topaz enlarging software A.I. Gigapizel which is
<<<supposed to be fantastic at making huge enlargements using artificial
<<intelligence. I'm impressed so far but it takes several hours to process
<<<one photo! I'll have to be very select with what I try!
Digitally Immobile Moose:
<I've tried it twice. First time, it locked up my computer. OK, so it was a
full size Raw file. So I tried a much smaller TIFF. I watched the green bar
move slowly across until it looked to be about finished - and BAM, computer
locked up solid. Reset, which I NEVER need, was the <only way out.
<They say it takes forever, but is supposed to run in the background. Not so
far.
<Ah, I see my GPU is one step below what's required. Problem identified, if not
<cured. :-(
Quite the resource intensive software but looks fascinating. Very glad Tina
mentioned it. I had used neural networks to model some complex physiologic
processes--had to use a separate expensive board as the 486 took hours vs
minutes for the optimized board. The network is only as good as the training
set used to construct it, however. It seems to do well with skin and at least
seascapes and suspect many others. See review too:
http://www.northlight-images.co.uk/topaz-ai-gigapixel-review/
The review implies it can be used as an image enhancing tool even if not used
to upres for a large print. The processed images then downsized for display
were improved. The software may actually "invent" details that were not there
based on network optimization gleaned from previous training sets unlike L-R
deconvolution that can only recover some hidden resolution/contrast generally
using Gaussian blur as the presumed PSF. A large crop processed through
GigaP then optimally processed by the usual tools might be superior. I wish
DIM would be mobile and try. May be especially good for MFT sensor with
fewer megapickles and might allow for more cropping with minimized loss of IQ,
if my hypothesis is correct. I am not sure would trust the converter in GigP
though.
Mike
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|