Hacker News new | ask | show | jobs
by piqufoh 2615 days ago
I suspect it's been generated from the generator part of a GAN (Generative Adversarial Network); you get two ML algorithms - one to pick real from fake pictures (a classifier) and one to generate fake pictures (the generator) - and then pit them against each other. I wonder if this image is possibly a high quality generative example that the classifier failed to identify as 'fake'.

Also - the generator probably uses an autoencoder to generate the images, which might explain the 'how' https://en.wikipedia.org/wiki/Autoencoder.

I hope that explanation wasn't too dull!