Microsoft files a patent for a new machine learning model that basically lets you construct hyper realistic avatars that look a lot more alive than they already do.
The new patent,
published on USPTO’s portal on November 14 this year, describes that the new machine learning model can make avatars and photos look like they are experiencing physiological processes, such as blood flow for example.
The method involves adjusting pixels based on a physiological signal, such as heart rate, to create a more natural appearance in videos or images.
But, why? Well, Microsoft says that when realistic avatars are too realistic, people find them creepy. So, to make them more “trustworthy” and lively, physiological signals — like blood flows or blushes for example — can be added to make them more lifelike.
The Redmond-based tech giant further describes in the document that realistic avatars need to mimic not just obvious movements like blinking or head gestures, but also subtle changes like variations in blood flow, breathing, and even reactions to emotions.
So to achieve this goal, Microsoft will use a deep learning model called a convolutional attention network (CAN) to improve the accuracy of capturing blood flow patterns on an avatar’s face.
This model then considers both color changes and temporal dynamics (changes over time) to create a more realistic representation of blood flow.
The application of this patent should go beyond further than daily use: it can also be applied to creating video game characters, etc. Maybe the new 3D avatars on Microsoft Teams will get a whole lot more realistic. Who knows, but one thing is for sure, if this patent goes through, then it would be a game changer that could revolutionize the way avatars and animated characters are perceived.
source
