Abstract
As avatar images become more photorealistic, users may start noticing fine-grained details (e.g., wrinkle movements) that significantly degrade photorealism. This disclosure describes techniques to generate photoreal avatars that can convey nuanced facial expressions. In contrast to traditional single-image-based avatar generation, with user permission, multiple enrollment images of a user with various facial expressions are obtained. The enrollment images are leveraged to model novel facial expressions in the avatar of the user. By following driving signals, the techniques can generate plausible avatar images with accurate facial expressions, including high-frequency details such as wrinkle movements. The techniques help in surpassing the uncanny valley that occurs with photorealistic avatars.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Zhang, Yinda; Pandey, Rohit; Fanello, Sean; and Tan, Feitong, "Improving Avatar Expressiveness with Feature Blending in Parametric Model Space", Technical Disclosure Commons, (November 28, 2024)
https://www.tdcommons.org/dpubs_series/7599