Encoder-Decoder Facial Geometry Encoding for Cross-System Compatibility
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face re-enactment techniques lack compatibility across different neural network implementations due to unique configurations of encoder and decoder neural networks, leading to difficulties in interoperation and increased data transfer volume.
Innovation Solution
The use of interpretable facial representations that disentangle geometric attributes from identity information, allowing for encoding and decoding geometric information separately from fundamental images, thereby enhancing compatibility and reducing data volume.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If face re-enactment techniques use unique encoder and decoder neural network configurations, then system-specific performance is improved, but compatibility across different implementations deteriorates
Solution Approach 1:
The patent segments facial representation into two independent components: geometric attributes (pose, facial movements) and identity information. By encoding geometric attributes separately using standardized parameters that are independent of specific neural network implementations, the system achieves both specialized performance and cross-system compatibility. The geometric information is extracted and encoded in a manner that can be universally interpreted across different encoder-decoder configurations.
2Loss of information
If all facial information is encoded together, then complete facial representation is achieved, but data transfer volume increases
Solution Approach 1:
The patent extracts geometric attribute information from the complete facial representation and separates it from identity information. By taking out only the essential geometric attributes (pose parameters, facial movement parameters) that are necessary for face re-enactment and encoding them separately, the system reduces the overall data transfer volume while maintaining complete facial representation capability. The extracted geometric information can be transmitted more efficiently than raw pixel data or fully encoded facial representations.
Data Source
AI summary
An encoder includes memory and circuitry coupled to the memory. Using the memory, the circuitry: encodes at least one fundamental image for use in displaying a video; and encodes, as information corresponding to each of images of the video, geometric information indicating geometric attributes within a region including a face of a person.