Microsoft's New AI: Virtual Humans Became Real! 🤯

TL;DR

Microsoft's AI can now accurately detect facial landmarks using virtual humans, improving on previous techniques and running in real time on any phone.

Transcript

Dear Fellow Scholars, this is Two Minute Papers with Dr. Károly Zsolnai-Fehér. Today, we are going to see Microsoft’s AI looking at a lot of people who don’t exist, and then, we will see that these virtual people can teach it something about real people. Now, through the power of computer graphics algorithms, we are able to create virtual worlds, a... Read More

Key Insights

• 🚂 Virtual humans provide a valuable source of data for training AI algorithms.
• 👶 The new technique significantly improves facial landmark detection by tracking over 700 landmarks.
• 🏃 The method shows improved consistency and can handle occlusions, running in real time on any phone.

Questions & Answers

Q: How does Microsoft's AI use virtual humans to improve facial landmark detection?

By using virtual humans, Microsoft's AI can generate perfectly annotated data, allowing it to train algorithms to detect facial landmarks with accuracy.

Q: What are the advantages of using virtual humans for training AI?

Virtual humans provide unlimited data that is perfectly annotated, eliminating issues of identity or distribution. They also allow for experimentation in different environments and wardrobes.

Q: How does the new method compare to previous techniques?

The new method tracks over 700 facial landmarks, improving on previous methods that could only track a fraction of that. It also shows improved consistency and can handle occlusions better.

Q: Can the new technique be applied to other applications, such as DeepFake videos?

Yes, variants of this technique may improve the fidelity of DeepFake videos by enabling better tracking of facial landmarks and democratizing the creation of virtual worlds.

Summary & Key Takeaways

• Microsoft's AI uses virtual humans to train and improve facial landmark detection algorithms.

• The new technique can track over 700 facial landmarks, improving on previous methods.

• It solves issues of temporal consistency and can handle occlusions, such as hair or clothing covering the face.