Me and Rafael Izbicki (PhD's at CMU, statistics & math - https://rizbicki.wordpress.com/) are trying to create an artificial intelligence capable of drawing and creating media autonomously.

1) In the drawing part, we started with Google paper's paradigm (https://arxiv.org/abs/1704.03477), replacing some key algorithms and making mathematical additions in order to allow the A.I. to make 3D photo-realistic drawings;

b) On the stylization side, we started with the Google paper's paradigm (https://arxiv.org/abs/1610.07629), replacing some equations and adding some insights as follows:

i) We used a recommendation system + RNN in order to customize our stylizations, taking into account either an image's classification and it's specific contents (i.e., which "filters" look better with landscape images, with specific rocks on a landscape,
etc.? And what about with people? With certain objects? and so on) and also preferences of the observer (this user prefers certain aesthetics?);

ii) Finally, we have allowed our stylizations to consider colors and shapes of each image in order to make a better style transfer (i.e., to make the curves of an image receive a stylization that also has curves, thus making them coincide);

We would like to know if one of you guys – an experienced developer – is interested in continuing this research with us, configuring the code environments, connecting our developed solutions and adding some other stuff, such as:

1) A drawing compiler;
2) Making tests and improving the performance & calibrating our algorithms;
3) Allowing the output to be exportable to editable formats;
4) After the A.I.'s drawings are concluded, allowing manual edits.

We are really committed to this side project – it will become a paper and perhaps something else –, so only respond if you have real interest :)

Thank you very much and I hope to hear back from you!

--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)