CVPR Daily - Tuesday

9 / 30

Phillip Isola is presenting his paper
“Image-to-Image Translation with
Conditional Adversarial Networks”,
which is joint work together with Jun-
Yan Zhu, Tinghui Zhou, and Alexei A.
Efros. Their idea is to use generative
adversarial networks (GANs) to solve
image-to-image mapping problems,
and in their paper they demonstrate
that these are a general-purpose tool
that can be applied to a lot of
problems.
GANs, which were introduced by Ian
Goodfellow et al. in 2014, and are a
popular idea at the moment, and a
large part of our community has
gotten quite excited about them -
“rightfully so”, Phillip says. He told us
that previously a lot of people have
done work on unconditional GANs,
which were used to generate random
images. But Phillip and his co-authors
thought that it might be more
compelling to look at the conditional
case, where you use a GAN for
regression problems to learn a
mapping from inputs X to outputs Y.
Phillip Isola 9
Phillip Isola is a postdoc with Alyosha
Efros at UC Berkeley.
Image-to-Image Translation with Conditional Adversarial Networks
“There’s a lot more problems
that are conditional than
unconditional, especially
practical problems in
computer vision and
graphics”