Deepfakes for dancing: you can now use AI to fake those dance moves you always wanted

Artificial intelligence has proved to be a very useful tool when manipulating people's videos. Face exchange depth fake is the most obvious example, but I am looking for new uses everyday. What is the latest? Let's call it a deep insight of dance. Use AI to read someone's dance movements and copy them to the body of the target.

The actual science here was done by four researchers at the University of California at Berkeley. As mentioned in the articles published at arXiV, their system consists of many separate steps. Initially, the target video is recorded and the subroutine converts the action to a stick figure. (You need a large amount of video to get high quality transmission, about 20 minutes of material, 120 frames per second.) Then find the bar graph consisting of the source video and their actions. Thereafter, an exchange is made and the neural network synthesizes the target individual's video based on the movement of the source bar graph.

Of course, this seems easy, but there are many clever projects here. For example, there is a subroutine to smooth the motion of the bar graph so that the dancer does not swing much, and there is a completely independent neural network dedicated to backtracking the target's face to ensure realism.

There are limits to the plan as well. For example, in a network, it is not possible to accurately simulate a loose fabric, so the target individual needs to wear tight clothing. In the video above, we also see a considerable number of visual anomalies, such as the moment when the target dancer and the source dancer are incompletely aligned, or the complex movement such as the front of the palm is turned over from the back can not be reproduced. Yes.

But this is a wonderful job. This video operation may take a day. Now, just need some source video and appropriate AI software. It is expected to become an application long ago and do not consider the impact that such technology has on our trust in video evidence.

Another big blow on the Internet today is the current trend of deep fake. Deepfakes is a fake porn video produced by artificial intelligence, and the face of porn star is exchanged with anyone (http://bit.do/df13). Deep fake usually replaces celebrity's face with erotic performers but people use them to make fake porn for their friends and classmates (http://bit.do/df11) . This is very worrying as people can use it as extortion mail. On social platforms where machine learning and continuous improvement of artificial intelligence and information dissemination can be promptly disseminated, it is often not possible to distinguish fake information from actual information. What I am paying attention to is that people do not know this question at all.

Artificial intelligence has proved to be a very useful tool when manipulating people's videos. Face exchange depth fake is the most obvious example, but I am looking for new uses everyday. What is the latest? Let's call it a deep insight of dance. Use AI to read someone's dance movements and copy them to the body of the target. The actual science here was done by four researchers at the University of California at Berkeley. As mentioned in the articles published at arXiV, their system consists of many separate steps. Initially, the target video is recorded and the subroutine converts the action to a stick figure. (You need a large amount of video to get high quality transmission, about 20 minutes of material, 120 frames per second.) Then find the bar graph consisting of the source video and their actions. Thereafter, an exchange is made and the neural network synthesizes the target individual's video based on the movement of the source bar graph.

Making celebrity porn photos is not new. However, at the end of 2017, Reddit 's Deepfakes began applying deep learning to create celebrity counterfeit video. This opens a new wave of fake video online. As a member of the US military, DARPA also funds research to discover fake videos. In fact, the application of AI to create videos before Deepfakes starts. Face 2 Face and UW 's Synthetic Obama (Synchronized from Lips of Voice Learning) created a fake video that is hard to find. In fact, they are so authentic, Jordan Peel has created a warning to the following general