Thanks to Sabrina Verhage I was pointed to the right direction to get some background to this project …
It was set up by Dutch Public Broadcaster NPO, and to cover the subject of neural network image translation system Pix2Pix they used as source material images of reporter Lara Rense:
A while ago we called on you to help us. Using listeners NPO Radio 1, site visitors of Knowledge Day and a few toiling hands on several editorial boards, we have managed to collect hundreds of drawings and photographs . All of the same object: News and Co- presenter Lara Rense. This program will create new drawings by Lara himself can change pictures of her. Gerda Bosman told about it in news and Co .
A true artificial intelligence system so: Based on the drawings and photographs data the system can invent how a picture based on a new drawing should look like. For example, he recognizes that your eyebrows are drawn and what should be Lara’s nose (see illustration above). But how does this artificial intelligence system exactly?
Tom Kenter, PhD Information Technology at the University of Amsterdam, explains. You carry a lot of pictures with accompanying drawings. This will calculate two networks started. One network is a “generator”, which calculates how a photograph of a drawing may look like. The result is evaluated by the “discriminator” who must say whether the image generator shows it is calculated by the algorithm, whether it is an original photograph.
So both networks hone their discernment. One takes pictures of drawings, the other assesses. If the generator does well - he makes a good picture of where the discriminator thinks it’s an original, the generator receives a reward and the discriminator a penalty.
dededos
kianamaiart
turnnoffyourmind
prguitarman
annaxmalina
