Exemplar GAN (ExGAN) are a type of conditional GAN that utilize exemplar information to reproduce quality personalized in-paintings. While GAN is a class of artificial intelligence algorithms in unsupervised machine learning, implemented through a system neural networks contesting with each other in a zero-sum game framework.

Facebook research attempt to replace closed eyes with open ones in a remarkably convincing manner through In-Painting Eye with Exemplar Generative Adversarial Networks. This is an instance of intelligent “in-painting” when a machine fills in a space with what it thinks should be at the point.

ExGANs successfully produced photo-realistic personalized in-painting results that are both perceptually and semantically plausible by applying them to the task of closed-to-open eye in-painting in natural pictures.

Facebook researchers were able to include “exemplar” data showing the target person with their eyes open, from which the GAN learns not just what eyes should go on the person, but how the eyes of this particular person are colored, and the actual shape.

The test results were stunning, as most people mistook the fake eyes-opened photos for real, and unless they knew a photo was definitely tampered with, they probably couldn’t differentiate it in the Newsfeed.

Albeit, the capabilities still falls short in some instances, resulting some strange looking artifacts of a person’s eye if partially covered by a lock of hair, and sometimes giving a complete color mismatch.

However, their detailed nature make it particularly useful as the capacity to convincingly replace the eyes weren't available in other photo editing tools. Facebook has introduced the new benchmark dataset for the task of eye in-painting for future comparisons.

Facebook demos Exemplar GAN (Generative Adversarial Networks) In-Painting to produce personalized results



Exemplar GAN (ExGAN) are a type of conditional GAN that utilize exemplar information to reproduce quality personalized in-paintings. While GAN is a class of artificial intelligence algorithms in unsupervised machine learning, implemented through a system neural networks contesting with each other in a zero-sum game framework.

Facebook research attempt to replace closed eyes with open ones in a remarkably convincing manner through In-Painting Eye with Exemplar Generative Adversarial Networks. This is an instance of intelligent “in-painting” when a machine fills in a space with what it thinks should be at the point.

ExGANs successfully produced photo-realistic personalized in-painting results that are both perceptually and semantically plausible by applying them to the task of closed-to-open eye in-painting in natural pictures.

Facebook researchers were able to include “exemplar” data showing the target person with their eyes open, from which the GAN learns not just what eyes should go on the person, but how the eyes of this particular person are colored, and the actual shape.

The test results were stunning, as most people mistook the fake eyes-opened photos for real, and unless they knew a photo was definitely tampered with, they probably couldn’t differentiate it in the Newsfeed.

Albeit, the capabilities still falls short in some instances, resulting some strange looking artifacts of a person’s eye if partially covered by a lock of hair, and sometimes giving a complete color mismatch.

However, their detailed nature make it particularly useful as the capacity to convincingly replace the eyes weren't available in other photo editing tools. Facebook has introduced the new benchmark dataset for the task of eye in-painting for future comparisons.

No comments