The trendy artificial intelligence technology is currently developing rapidly, and at the same time more and more artificial intelligence applications are beginning to appear in our lives.
Previously, developers used Google’s open source free convolutional neural network tool to replace characters in pornographic films with stars and achieved a rich and vivid point.
Recently, researchers at the University of California, Berkeley, used artificial intelligence to build models to help users automatically generate dynamic dance videos.
Let’s take a look at the video released by the researchers:
Split different actions and artificial intelligence to generate a coherent picture:
This application initially requires the collection of professional dancers to draw motion models. From the video, it can be seen that the dancer’s movements are simplified for artificial intelligence to learn.
Then artificial intelligence can generate ordinary user-generated actions according to the model according to the learning model, and finally complete the dance fragments that are almost coherent and fake.
The models that the research team has produced now include a variety of dances, which rely on different neural network modules to be responsible for different tasks for a successful generation.
For example, subsystem A analyzes the action of the professional dancer and then draws the action model, and subsystem B tracks the user’s face to ensure realism, and finally draws the action.
However, it is currently not possible to handle too complex actions:
Researchers say that this artificial intelligence application can basically complete dance generation, but if you look closely, you will find some fragments.
This is because artificial intelligence is temporarily unable to accurately simulate loose clothing, so the target individual must wear tights to ensure that the segments are not blurred.
Of course, there is some complicated motion artificial bits of intelligence that can’t be used for it. For example, when the hand moves backwards and forwards, the motion of the hand is occluded and folded.
Still, it seems that this is still a very interesting artificial intelligence application, and perhaps more and more videos can be forged in the future.