We can easily say the progress we see now. Artificial intelligence Our working relationships and many of our jobs will change in the coming years. we are in the space race artificial intelligence and all the big companies technology They are improving their systems. For example Meta, from Google and Microsoft.
systems artificial intelligence most researched viewing. In other words, if we textally describe an image in our head to such systems, artificial intelligence it will quickly produce different variations of what we want from it.
there is also artificial intelligence video production they recently presented their system which works the same as the previous one but instead of rendering images it will create video from the text we want from the system. artificial intelligence voice translation and others that “revive the voice” of someone who has died from a set of data from the deceased.
Style transfer is a technique based on artificial intelligence A photograph in which a photograph is taken, with a table next to it, and the style of the photograph is applied.
On the other hand, pose estimation is a technique where we have a photo or video of someone and the output is a skeleton of that person’s current stance.
So how about something that combines the power of pose prediction with the expressive power of style transfer?
As can be seen from the video, we can shoot a video of a professional dancer and shoot our own video, let’s say moderately beautiful movements, and then we can transfer the dancer’s performance to our own body in the video. scientists call it motion transmission.
The videos they use as examples are pretty fluid, and that’s no coincidence. With this technique, time consistency is taken into account. This means that the algorithm knows what it’s doing right away and won’t do anything too different, which makes the moves in this dance smooth and believable.
This method uses a generative antagonistic network where we have a neural network to predict the pose, or in other words, we construct a skeleton for an image and create a generative network to generate new images when a test subject and a new pose are given. skeleton.
These two neural networks work with each other, teaching each other to create and distinguish increasingly unique images over time.
Some artifacts are still out there, but keep in mind that this is one of the first articles on the subject and is already in incredibly good shape. It’s fresh and experimental.
In short, this motion transfer based artificial intelligence is called “do what I doFrom a video of people dancing, this performance can be transferred to a new (amateur) target just minutes after the target subject performs standard movements.
This issue was addressed as a video-to-video translation where exposure was used as an intermediate representation. To convey motion, poses were extracted from the source subject and learned mapping from pose-to-appearance was applied to create the target subject.
#Motion #transfer #based #Artificial #Intelligence #latest #technology