Microsoft Granted Patent for Deep Learning-Based Multimodal 3D Facial Modeling

April 24, 2024 – According to a recent listing by the United States Patent and Trademark Office (USPTO), Microsoft has been granted a new patent for creating highly realistic virtual portraits through deep learning-based multimodal 3D facial modeling.

The patent, titled “Multimodal Three-Dimensional Facial Modeling and Tracking for Generating Expressive Virtual Portraits,” outlines a system that utilizes a processor and storage system. This processor is tasked with receiving initialization data, which represents the starting appearance of the facial model, along with additional multimodal data signals such as audio data from the user’s voice.

The technology employs 3D facial modeling and tracking techniques to generate 3D vertices based on the user’s face. These vertices are then transformed from a neutral facial expression to depict various emotions on the digital facial model.

Furthermore, the system employs deep learning to design virtual portraits that can mimic the user’s facial expressions. Users also have the ability to fine-tune their virtual portraits for further personalization.

This innovative patent from Microsoft paves the way for more realistic and expressive virtual representations, potentially enhancing various applications such as gaming, movies, advertising, and more.

Leave a Reply