1
$\begingroup$

Simplified: What is it called in AI when a program is designed to make "x in the style of y;" when it trains off of two types of sources in order to make a thing from source one, informed by features from source two? For example, if a network made up of two smaller networks were to take sheet music of a specific compositional style in network A and audio samples from a certain genre in B and through an interface creates music from a certain genre in a certain compositional style; the is comes from One, the seems comes from Two.

For more coarse and obvious examples:

  • "Compose synthpop in the style of Beethoven"
  • "Draw impressionism in the style of Mondrian"
  • "Generate casserole recipes using only ingredients most likely to fluctuate in price given current market data"
  • "Sketch baseballs that look like they're made of espresso foam"
$\endgroup$
1
2
$\begingroup$

This problem is known as "style transfer". The field was started by Gatsys et al. in 2016, and has seen a lot of work over the last few years, including conditional translation from Isola et al. in 2017 and unpaired translation from Zhu et al. 2017.

This picture (Figure 2 from Gatsys et al.) shows the idea behind style transfer, and illustrates the dramatic results that are possible with these models.

enter image description here

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.