Facial Geometry Encoding for Compatible Low-Bandwidth Video Re-Enactment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face re-enactment techniques lack compatibility across different neural network implementations due to unique configurations, leading to difficulties in interoperation and inefficient data transfer.
Innovation Solution
Encoding geometric information, such as feature point locations, instead of entire images, to facilitate universal reuse and reduce data volume, while ensuring natural output video generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If entire images are encoded for face re-enactment, then image quality is preserved, but data transfer volume increases
Solution Approach 1:
The patent extracts only the essential geometric information (feature point locations, facial landmarks, pose parameters) from complete images. Instead of encoding entire image data, the system identifies and encodes only the critical geometric attributes that define facial geometry, thereby reducing data volume while preserving the necessary information for face re-enactment.
Solution Approach 2:
The patent segments the face into discrete feature points and geometric elements. By dividing the continuous image data into distinct geometric components (landmarks, contours, pose parameters), the system enables selective encoding of only the essential geometric information needed for reconstruction, rather than transmitting complete image data.
2Reliability
If different neural network configurations are used, then model specialization improves, but compatibility across implementations deteriorates
Solution Approach 1:
The patent establishes a universal geometric information encoding format that can be used across different neural network implementations. By defining a standardized structure for representing facial geometry (feature points, landmarks, pose parameters), the system enables compatibility between different models while allowing each implementation to specialize in its own processing approach.
Solution Approach 2:
The patent changes the representation parameters from raw image pixels to standardized geometric attributes. This parameter transformation creates a model-agnostic intermediate representation that can be processed by various neural networks while maintaining consistency in how facial geometry is described and transmitted.
3Quantity of substance
If geometric information is encoded instead of entire images, then data transfer volume reduces, but information completeness may be compromised
Solution Approach 1:
The patent applies partial action by encoding only the essential geometric information needed for face re-enactment rather than complete image data. By identifying and encoding only the critical feature points and geometric attributes that define facial geometry, the system achieves sufficient information completeness for the specific application while significantly reducing data volume.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An encoder (100) includes memory (152) and circuitry (151) coupled to the memory (152). Using the memory (152), the circuitry (151): 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.