2017.09.27 Up

Posted by Nao Tokui

変化し続ける架空の人物ポートレート – Portraits of Imaginary People

変化し続ける架空の人物ポートレート – Portraits of Imaginary People

9月のArs Electronicaでも展示されていたMike Tykaさんの作品. GANを用いて、ポートレート(肖像画)が変化し続けていく作品です.



まずは一般的なDCGANで128 x 128の荒い絵を生成したのちに、目や口といったポイントとなる顔の要素を識別・切り出し、それらを解像度をあげるGAN (Superresolution)にかけます. これで32 x 32の顔のパーツが、256 x 256にアップサンプリングされます.  ポートレートという領域に限定していることで可能になる手法ですね。





Portraits of Imaginary People explores the latent space of human faces by training an artificial neural network to imagine and generate portraits of non-existent people.
To do so, thousands of photos of faces from Flickr were fed to a type of neural network technique called a “generative adversarial network” (GAN). GANs work by using two neural networks playing an adversarial game: one (the “generator”) tries to generate increasingly convincing output, while a the second (the “critic”) tries to learn to distinguish real photos from generated ones.
At first both networks are poor at their respective tasks. But as the discriminator network starts to learn to predict fake from real, it keeps the generator on its toes, pushing it to generate harder and harder examples. As the generator gets better the discriminator also has to improve in turn, in order to keep up. With time, the generated output becomes increasingly realistic, as both adversaries try to outwit each other.