Facebook’s AI analyzes your look and helps you with dressing.
Facebook presented several artificial intelligence projects at the International Conference on Computer Vision 2019. Several of these projects exploit computer vision technology in an original and unexpected way…
As part of the International Conference on Computer Vision 2019, Facebook recently presented many artificial intelligence projects. These projects include Machine Learning models designed to anonymize faces, simulate hand movements or modify people’s clothes in video.
This concept is reminiscent of Deepfakes, which aims to integrate a person’s face into a video using Machine Learning. However, Facebook researchers are looking to use this technology in a positive way.
Thus, just like Deepfakes, Facebook’s AI maps a person’s facial expressions and movements based on their different characteristics. However, rather than inverting the person’s face with that of another, Facebook simply changes the face until it is no longer possible to recognize it for facial recognition systems.
This technology could allow those who wish to remain anonymous to express themselves in the video. This system will need further optimization before it can be deployed as a product, but it could prove very useful in the near future when facial recognition will be able to identify anyone instantly.
Facebook’s AI simulates hand movements and relocks you
The second study published by Facebook presents an AI capable of capturing, cataloging and reproducing body language movements. More precisely, the system is currently focusing on hand gestures.
To train this AI, the researchers recorded 50 hours of video during which pairs of people equipped with motion sensors conduct ordinary and natural conversations. These videos were then transferred to the Machine Learning model to allow it to associate gestures with expressions. For example, the system may have noticed that a person will often point behind them when talking about the past or that they will usually make a sweeping gesture when saying “anywhere”.
This technology can be used in particular in the field of virtual reality. Virtual avatars can learn to mimic users’ hand movements. Facebook could, therefore, use this innovation to increase the realism and immersion of its social network in VR “Horizon” expected in 2020.
Another AI presented by Facebook at the conference is designed to help you dress better. Called “Fashion++”, the system was trained from a large library of images labeled to indicate both the type of clothing worn by the subject and the level of “style” of his costume.
Now, Fashion+++ is able to analyze an outfit in real time to suggest changes. The AI is not yet advanced enough to make you look like Cristina Cordula, but enough to advise you to remove a thickness or tuck in your shirt.
This is only the beginning, and this artificial intelligence could eventually become a true fashion expert. Such technology could be integrated into future connected mirrors, to provide advice to users and help them dress better…
In general, the various projects presented by Facebook to the ICCV show that researchers are exploring all the possibilities offered by computer vision. Whether or not these researches are translated into commercial projects, they prove that we have only explored a tiny part of the applications of artificial intelligence…