Neural Network Transformations

[Chris Kern]

As I mentioned elsewhere on this site a while back, I've been experimenting on-and-off lately with using convolutional neural networks to alter photographs into something that no longer qualifies as genuinely photographic but also isn't traditional manual graphic art.  I call these derivative images transformations, for want of a better term.  I also think of them as a collaboration between me and the software―a "collaboration" because neither I nor the software completely controls the result.  I've attached a recent sample which this evening passed an important test: my 16-year-old granddaughter, who has developed a (very welcome) interest in photography, pronounced it "really cool."  Which emboldened me to submit it to this esteemed, if ... ahem ... ever-so-slightly more mature forum.

Attached: (1) the original photograph and (2) the transformed image, Blowing Bubbles in Shinjuku Chuo Park, Tokyo, Japan, a composite of two neural network outputs that I combined in Adobe Photoshop (photograph: 2017, transformations: 2019).

[armand]

I wouldn't call this a photograph anymore but the result is striking, I like it. I don't know if I would still like it if I start seeing this kind too often but here works quite well.

Edit. some grammar correction, when typing in a rush on the phone things don’t go so well

[Rob C]

Well, we have become accustomed to plastic models in portraits and fashion photos, so why not woodcuts too?

:-)