Logo Synthesis and Manipulation with Clustered Generative Adversarial Networks (opens in new tab)

Nipple? Apik? Could be a sportswatch. There might be a few clusters along the path through latent space (if that makes any sense, just skimmed the paper mostly for the figures).

The attempt here seems to be really naive, I agree. But why are logos not compositing? Coat of arms are frequently described in such a manner that would allow to mix them. But then, the traditional artistic combinations of different ones into new are not mere half way morphs. And a classic logo needs to be compositional, because it's easier to perceive (decompose), e.g. hammer and sickle. Scientific Icons are frequently using mathematical patterns and plots, which tickle the eye in quite a different manner. I thought the nike swoosh comes from that rather abstract direction, whereas the apple is quite objective. Both are pictographs, but only the apple is a logo (from logos, ie. speaking).

Hi, I'm one of the authors. That is correct in a strict sense, but we wanted to focus on the more 'creative' part of logos rather than the text. GANs are known to struggle with high resolutions, but we note that we show higher resolution logos later in the paper (see page 12 and 15 e.g) which is trained on the smaller but higher-resolution version of our dataset.

I found interesting the identification of different clusters. Now do some learning over the clusters. Letter cut out from colored shape seems to be the most prominent feature, and the most boring one. But some of the shapes (without cut-outs) are rather interesting. Some of the figures from he article could become icons themselves, reminiscent of scientific plots (cf. https://upload.wikimedia.org/wikipedia/commons/thumb/b/b0/Un...).

The problem is, a logo should be as unique as possible, so mechanical derivatives aren't convincing