Back
Yiannis Kompatsiaris (CERTH-ITI)
Author/s
About the resource/s
Generative Adversarial Networks (GANs) are part of the cutting edge in recent machine learning research. Trained by exploiting the interplay between a generative and a discriminative deep neural network, GANs are able to implicitly model the data distribution, thus allowing us to generate realistic data samples after model training is complete. The Digital Media sector is perhaps the greatest beneficiary of GAN-based algorithms, with industry-relevant research already taking advantage of their potential. They are useful for a wide range of media-related applications, such as image-to-image translation, optical recognition enhancement, text-to-speech systems, music/image/video generation, video summarization, image anonymization and many others, frequently giving rise to state-of-the-art results. On the negative side, the rise of so-called “deep fakes” poses significant new challanges, both ethical and technical, thus making the development of methods to robustly detect them a top priority. Overall, GANs hold the key to revolutionizing the way we produce media and arts content.
Media
