Image Source (here)
Over a few years, applications of the Generative Adversarial Networks (GANs) have seen astounding growth. The technique has been successfully used for high-fidelity natural image synthesis, data augmentation tasks, improving image compressions, and more. From emoting super-realistic expressions to exploring deep space, and from bridging the human-machine empathetic disconnect to introducing new art forms, GANs have it all covered.
Here, we list down a few impressive real-world applications of GANs.
(The list is in no particular order)
1| Bringing Monalisa Back To Life
Who imagined that one day we will be able to see the expressions of the Italian noblewoman Lisa Gherardini from the famous portrait of Mona Lisa painted by the Italian artist Leonardo da Vinci. Yes, this has made possible by the advent of DeepFake.
A team of researchers at Samsung AI, Moscow, used a machine learning system that includes Few-Shot adversarial learning to create expressions in the portrait. The system performs a lengthy meta-learning on a large dataset of videos and after was able to frame a few one-shot learning of neural talking head models of previously unseen people as adversarial training problems with high capacity generators and discriminators.
2| Showing Artistic Skills
GAN has shown its capability of creating impressive portraits from scratch. There are several instances where GAN had shown its to-the-point artistic skills. For instance, the Art of Mario Klingemann was auctioned at the Sotheby’s Contemporary Art Day Auction.
Mario Klingemann created a machine learning system known as Memories of Passersby I that uses neural networks in order to generate an infinite flow of portraits. Mario trained the system with the help of more than a thousand portraits dated from the 17th to 19th centuries.
3| Creating Deepfake Videos
Deepfake is becoming one of the most discussed topics for researchers when it comes to safety and security. Developers have been using deep learning technology for generating fake faces and impersonating others for a few years now.
In a blog post, Jerome Pesenti, who leads the development of AI at Facebook, stated that deepfakes are an important concern, so they developed a system to identify any menace. There have been a few instances of deepfake such as the above video where Cardi B’s face had been changed into Will Smith’s.
Last year, we also witnessed the video where a developer deepfaked Jim Carrey into Nicholson’s most popular cult classic “The Shining”. The creator of this video used a commonly available open-source tool known as DeepFaceLab. DeepFaceLab has become the goto application for developers to create deepfakes in a quick and easy manner.
4| Making Photos Better
With the help of GAN, a team of researchers at the social media giant, Facebook created an approach known as Exemplar Generative Adversarial Networks (ExGANs) that can produce photo-realistic personalised in-painting results that are not only perceptual but also semantically plausible. This approach can be applied to the task of closing and opening eye in-painting in natural pictures. In simple words, this GAN approach can help you open a pair of closed eyes in a photo.
In the process, the generator network in the GAN masters to fill in the missing regions of a given image while the discriminator network learns to judge the difference between both in-painted and real images. This, in result, forces the generator to produce in-painted results that smoothly transition into the original photograph.
5| Swapping Faces
Deepfake has indeed travelled a long way in the domain of technology. Last year, AI-powered photo manipulation and editing app FaceApp made waves on the internet. The app is said to be the most advanced neural portrait editing technology which allows its users to make the person on the photo look younger or aged.
After FaceApp, another Deepfake face-swapping application, ZAO created a buzz that provides a number of features on photography for Android smartphones. The app also allows the users to add their faces on pre-defined clips.
Face swapping is used to transfer a face from an image source to a target image while face reenacting or face puppeteering uses the facial movements and expression deformations of a control face in one video to guide the motions and deformations of a face which is appearing in another video. Researchers from the Bar-Ilan University and the Open University of Israel have developed a similar model known as FSGAN for face swapping and re-enactment in images and videos.
6| Making A Pizza
Culinary arts can be said as one of the complex challenges for an intelligent system for building something sensible out of raw inputs. Last year, a team of researchers from MIT and Qatar Computing Research Institute worked on a machine learning system which can follow a recipe and make a pizza.
In this research, to achieve a system that can perceive food making as following a manual, the researchers composed operators that can add or remove ingredients from a dish, and here, each of the operators is actually a GAN which predicts how the food looks after every step.