Facebook is eyeing new artificial intelligence research - @omr.dgli

A new artificial intelligence research conducted by Facebook places eye on the faces of people who are closed eyes. With this technique in the name of internal painting, the program fills the gaps with what it thinks it belongs to.

In fact, we have already seen a similar system applied by Adobe in part. In Adobe Photoshop facing users with the content-aware fill setting, Adobe provided the photo with unwanted items filled in with other items in the background. For example, an unwanted electrical post that appeared in the photo could be destroyed by replacing it with clouds in the background.

However, this setting can not be said to be a solution in every case. This includes the eyes of the people in the photo. Nevertheless, eyes are so characteristic and original that they can not be re-designed from scratch. In this sense, we can say that the new research of Facebook responds to this need.

The source of this new research of Facebook benefits from the Generative Adversarial Network, a machine learning system. For those who are not familiar with GAN, let's turn this machine learning system a bit. In the GAN, the producer network constantly learns to produce new data, while the working differential network learns to discriminate between the data set accepted as input and the data generated. So, in this process, both networks learn about what to produce and what to distinguish irregularly on their own.

To give an example of this in Facebook; While one part of the system learns faces, the other part learns to create similar visuals. Under this feature, the network is working on getting to know both open eyes and mimicking them. In the early stages of this work, artificial intelligence directly affixed the clear eyes of the illustrative photographs to the photographs of which the eyes were closed.

So artificial intelligence, at this stage, passed through factors such as skin color, light, pose and eyes positioning in the eyes closed eyes. At this point, Facebook researchers created sample images that show their targeted faces while their eyes are open.

Since the network recognizes how the person looks, the color inconsistencies and the copy-and-paste feel are lifted to a more lasting appearance. It is stated that many people have difficulties in distinguishing eyes that are placed on the face and do not belong to that person. In short, this new network of Facebook is capable of doing its job skillfully. Still, the network continues to make mistakes in some cases. If someone's eyes are partially covered by their hair, different results can come out.

Although this new study is intriguing and frightening, it is unclear to what extent Facebook will use the result of this work. After a while, it is probable that Facebook will be in the same eye as the labeling proposal in the photographs of the eyes closed.

Steemit: @omr.dgli