This looks interesting, but the results are somewhat blurry.<p>Has anybody ever tried to use features of the logos (number of shapes, shape size, position, color, curvature, shape parents/children, etc.) instead of raw pixel data to train GANs?<p>Here comes the shameless plug :<p>ImageTracer is a simple raster image tracer and vectorizer that outputs SVG, 100% free, Public Domain.<p>Available in JavaScript (works both in the browser and with Node.js), "desktop" Java and "Android" Java:<p><a href="https://github.com/jankovicsandras/imagetracerjs" rel="nofollow">https://github.com/jankovicsandras/imagetracerjs</a><p><a href="https://github.com/jankovicsandras/imagetracerjava" rel="nofollow">https://github.com/jankovicsandras/imagetracerjava</a><p><a href="https://github.com/jankovicsandras/imagetracerandroid" rel="nofollow">https://github.com/jankovicsandras/imagetracerandroid</a>
I think logos are a tough problem for convnets because they're not very compositional - ie. they're not made of heirarchically nested parts.<p>the space of logos is also probably not continuous - eg. there is a logo in the latent space between nike and apple, but it's unlikely to be aesthetic.
These are iconographs, in the strict sense, not the full logos. The different google Gs don't really speak for themselves. The Y combinator Y is really not distinctive, either. The first few figures show fav-<i>icons</i>, I'd thought.
This title should be changed to "Smudge Synthesis". Move along nothing to see here. Actually the dataset of 600k logos is probably interesting. I bet someone who had some time could do a hugely better job.