Bitstream Model Signaling for Synchronized AI Decoding Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional audio and video compression technologies fail to adapt to long-term evolution and performance optimization due to static AI models, leading to suboptimal coding standards.
Innovation Solution
A method for encoding and decoding that includes setting a preset identifier value in a bitstream to indicate a second device to update or switch decoding and encoding models, allowing for synchronized model updates or switches, and optionally transmitting model information or index information to facilitate compatibility and reduce bandwidth requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If AI models are introduced to audio and video encoding and decoding, then coding performance is improved, but the coding standard cannot be updated or iterated in the long term
Solution Approach 1:
The patent introduces dynamic model switching capability where the decoding model can be switched between different versions (e.g., first decoding model and second decoding model) based on feedback information. This allows the coding standard to evolve and adapt to new AI models while maintaining backward compatibility, resolving the contradiction between improved coding performance and long-term standard evolution capability.
Solution Approach 2:
The patent implements a feedback mechanism where the first electronic device sends feedback information to the second electronic device about the encoding model used. This feedback loop enables the decoding side to adjust and update its model accordingly, facilitating long-term evolution of the coding standard while maintaining optimal coding performance.
2Manufacturing precision
If model information is transmitted to enable synchronized model updates, then coding performance is optimized, but bit rate overhead increases
Solution Approach 1:
The patent extracts only the essential feedback information (such as model identifier or version information) rather than transmitting the complete model data. This selective transmission approach optimizes coding performance through synchronized model updates while minimizing the bit rate overhead by sending only the necessary control information.
Solution Approach 2:
The patent uses a copying approach where instead of transmitting the entire AI model, only the necessary model parameters or identifiers are exchanged. The full model can be obtained through copying from a shared repository or cloud service, reducing the immediate bandwidth requirement while enabling performance optimization.
3Stability of the object's composition
If existing fields in the bitstream are reused as model identifiers, then backward compatibility is maintained, but new functionality must be integrated into existing structures
Solution Approach 1:
The patent applies multi-functionality by designing the model identifier field to serve multiple purposes: it can indicate different types of information (e.g., model version, model type, or update flag) depending on the context. This allows the same field structure to maintain backward compatibility while supporting new functionality for model updates and iterations.
Solution Approach 2:
The patent segments the model identification mechanism into separate components, such as using a model version field combined with a model type field. This segmentation allows existing fields to be reused for backward compatibility while adding new fields or parameters for updated functionality, reducing integration complexity by modularizing the approach.
Data Source
Figure 1A
Figure 1B(1)~1B(2)
Figure 1C(1)~2
AI summary
Embodiments of this application provide a decoding method and an electronic device. The method includes: first receiving a first bitstream, where the first bitstream includes a model identifier and a second bitstream, and the second bitstream is generated based on an encoded signal; obtaining model information when a value of the model identifier is a preset identifier value; then performing model reconstruction based on the model information, to obtain a decoding model; and decoding the second bitstream based on the decoding model, to obtain a reconstructed signal. In this way, a decoding model at a decoder side and an encoding model at an encoder side are synchronously updated, or a decoding model used for decoding at a decoder side and an encoding model used for encoding at an encoder side are synchronously switched.