Post by PoutnikPost by DieselPost by PoutnikYou mean the OP, as I do not have to.
I have just pointed out it was not for Windows.
It's not native Windows software, no. But, that can be worked around,
if one is willing to put forth the effort.
Sure, in the age of double booting and virtual machines.
Here's the tool I decided to use.
https://pypi.python.org/pypi/autocolorize
Relies on caffe at least. Caffe with Python3 bindings needs
a separate .ppa. Caffe with Python2 might be in your repository.
I installed the following, and almost got it working. The
color palette didn't show up, but the neural network loaded OK.
Something broke part way through, and because there were no debug
messages worth a damn on the screen with regard to "icolor"
window, I couldn't figure out where the lab_gamut comes from,
and what I needed to fix. But at least this encouraged me to get
set up with all the junk (scilab, numpy, caffe) needed, so I was
mostly familiar with what the one above was installing.
https://github.com/junyanz/interactive-deep-colorization
The trick with all this stuff, is making sure all the packages
you install, have Python2 or Python3 bindings. I would get
almost there, and realize "damn, this one is missing Python3"
and so on. It was a royal pain making sure the separately
installed packages all lined up. That's because the entire
project isn't in a repository, with a nice "meta install"
for it to drag in the dependencies.
The interactive-deep-colorization page has sample images in the
download. And that's what I used for this run of the
Autocolorize software.
The command used (I have caffe_cpu installed, as I have no GPU
worth using for Deep Neural Network usage) is below. I made a copy
of the parrot file, so it wouldn't get deleted by accident, and
that's what parrot2 is.
/home/paul/.local/bin/autocolorize -d cpu ~/Downloads/ideepcolor/test_imgs/parrot2.jpg -o ~/Downloads/ideepcolor/test_imgs/out2.jpg
And this is the comparison, source on the top, *auto* colorized
result on the bottom.
Loading Image...Notice that it wasn't able to "color within the lines".
The source photo really didn't look all that bad, in
terms of encouraging that kind of result. The orange
seems to have leaked out.
The hardest part of doing this, was finding the
energy to try it :-) Lazy, lazy, lazy.
Could you do that in Windows ? Uh, probably. There are
bits and pieces for Windows, but it was such a PITA to
get the Linux version running, I really don't think I
have the energy. If you find a website coloring things
automatically, it just might be using the above software.
The autocolorize program will download its own data files,
about 500MB worth. The files might not be exactly the
same as the interactive-deep-colorization ones. I haven't
bothered doing any forensics on those files (yet). Like,
why weren't they compressed, to save on downloads.
Now, isn't doing *zero painting* worth it ?
The results look damn good, for zero mouse clicks.
It's all done with a single command line invocation
(and a long delay as the 6GB of neural network stuff
is loaded). The neural network was trained on images,
to help it color within the lines, and recognize textured
areas. The messages printed on the screen when caffe loads
that, hints at what factors were used to build the
neural network.
If you had a modern video card of the right flavor,
that would run a lot faster than my conversion did.
It's a snoozer on a CPU. Caffe seemed to be running
on one core, and I couldn't find an option to use
more cores.
Paul