News for hackers. Stuff that matters.
Re:Imaginary worlds dreamed by BigGAN by minimaxir on Saturday November 17, 2018 @12:42AM (#18474164)
You can now demo BigGAN in an official Colaboratory notebook (backed by a GPU) to create your own AI-generated nightmares: https://colab.research.google.com/github/tensorflow/hub/blob...

Examples from the notebook by the article author: https://twitter.com/JanelleCShane/status/1062067001504321536

Re:Imaginary worlds dreamed by BigGAN by saganus on Saturday November 17, 2018 @11:35AM (#18476148)
Can someone explain to a layman what does the truncation parameter do?
Re:Imaginary worlds dreamed by BigGAN by Fr0styMatt88 on Monday November 19, 2018 @04:46PM (#18490008)
My understanding is that the 'truncation' parameter makes the output of BigGAN more closely resemble the training data set.

Moving the truncation towards 0 will give you more realistic but less varied images, while moving it towards 1 will give you more varied but increasingly potentially nonsensical images.

Re:Imaginary worlds dreamed by BigGAN by saganus on Tuesday November 20, 2018 @12:04PM (#18495404)
Aha!

That makes sense. Thanks!