Generating facial expressions using a neural network having layers of constrained outputs
US12073501B1 · kind B1 · utility
1Cited by
0References
20Claims
0Family size
Assignee
Inventor
Key dates
| Filing date | May 13, 2022 |
| Grant date | Aug 27, 2024 |
| Priority date | — |
| Expiry date | Oct 1, 2042 |
Classification
- Technology area (CPC G)Physics
- CPC primaryG06T2210/36
- WIPO fieldComputer technology
- WIPO sectorElectrical engineering
Abstract
Generating an expressive avatar includes obtaining an expressive image of a face, and determining at least one target levels of detail for a representation of the face. A trained convolutional neural network model is applied to the expressive image of the face to obtain a mesh representation of the face at each of the at least one target levels of detail, wherein each convolution in the trained convolutional neural network model provides a unique level of detail, and an avatar is generated based on the mesh representation of the face at least of the at least one target levels of detail.
Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.