Research coming out of Stanford University shows an interesting application of AI – generating fake tennis matches. Want to see a hyper-realistic tennis match between Roger Federer and Serena Williams, well now its possible!

So what does the AI actually do? Well, the first step is to study real-life footage to understand how the game of tennis is actually played.

How does the ball move? How do players hit the ball? How do the players move or stand while waiting for the opponent’s hit? All these things can be studied for the game as a whole, but also adjusted in style for each player.

The team boasts the AI is not only able to recreate a Tennis match which looks very realistic, but it will also take into account the two chosen players (even if they are both the same one) and will adjust according to their playstyle.

First, the AI predicts the path of the outgoing hit based on the hitting player & opponent. Next, it predicts the recovery predicts the return position of the hit (assuming the opponent does return it), and given that, where the player should place himself. The AI is based on the joint movement, only computing the movement for the skeleton itself. It is likely another AI (most likely GAN) that will realistically render a body over those joints.

With some refinement such as improving the surrounding (have the crowd/referee move) and include the shadows etc, the game would actually look very realistic. Potentially, it could change how sport-based realism computer games are made. Instead of making the flexible game-engine look as close as possible to the real footage, make the real footage flexible enough to act as a game engine!

You can fine the research paper here.

Title of the paper: “Vid2Player: Controllable Video Sprites that Behave and Appear like Professional Tennis Players”

Authors of the paper: Haotian Zhang and Cristobal Sciutto and Maneesh Agrawala and Kayvon Fatahalian

Disclaimer: I am in no way related to the authors of this paper or linked videos. I did not receive any compensation for this post.