Sailors and Soldiers Memorial Arch, Grand Army Plaza, Brooklyn, an edifice not far from my home, which started rising above the trolley tracks in October, 1892 and was topped off with Lady Columbia in December, 1898. The soldiers and sailors (other side) didn't arrive until April, 1901, finishing the piece. It's Brooklyn's answer to the
Arc de Triomphe, not that the Arc de Triomphe needed an answer, or even had asked for one, but such procedural trivialties have never stopped Brooklynites before.
Pulled from git after the Monday 11:46 (CET +02:00) commit, so I had the X event flush fix.
Pretty much effortless. I zoomed and panned and tried to break stuff. I found that - generally - only one hint dot is needed for each contiguous region (fore/background), so one background dot for the big sky and under the arch, one foreground dot for the arch, carved out about 95% of the mask needed. What kicked the dot count up to around 15 or so were all of the little discontinuous regions which Lady Columbia and her party carved out of the sky, so each little sky patch under the chariot and figures needed to be designated as background. The other spot that ate up a lot of hint dots is the sliver of sunlight falling in on the left hand side of the inner arch. The pale yellow of the stone is chromatically close to the pale greenish cyan of the sky, so a six or seven zig-zag of "This is the arch! (foreground)", "This is the sky! (background)" were needed before the local figure/ground relationship straightened out. I made this mask at 1X, which is not my usual habit; I usually like to work with 4X - 8X images, but this machine is for experimental/dangerous code and is a bit memory constrained at the moment. A larger image, and one, perhaps, that is actually an auxiliary "helper image," chromatically supersaturated via G'MIC's -orientation command, perhaps (chroma only; flattened luminance and saturation), would have permitted the creation of a mask with fewer hints. Maybe. That is something to play with. I also, (for what its worth), used the gmic command tool and -x_segment directly, which launches pretty much the same workflow as that in gimp-gmic plug-in land. When working with gmic directly, however, I have to tie a string around my finger and remember to normalize the alpha-mask to 0,255 before dumping off the image list, but that's a standard gmic thing.
Workflow comment. If this tool didn't exist, I'd have made a high contrast luminance mask and would have been retouching it a bit. Maybe twenty minutes or a half hour. I think here, I had the starting point for a good mask in easily five minutes. Allow, say, a generous ten minutes for post-production tweeking. So I think I'm out of the gate in half the time.
For something that was termed an "unmagical (dumb) algorithm," on the Google+ announcement, it seems pretty slick.
To celebrate, maybe I'll send you this little bit of Brooklyn bombasity as a postcard. I think its about the right size.
GRO
Login
Register
Tweets
FAQ
Advanced Search
Video from : 
