Bitstream Model Signaling for Synchronized Codec AI Updates
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing audio and video coding standards are not updated or iterated frequently, hindering long-term evolution and performance optimization due to the static nature of AI models used in encoding and decoding processes.
Innovation Solution
An encoding method that includes setting a preset identifier value in a bitstream to indicate a second electronic device to obtain model information for synchronously updating or switching encoding and decoding models, allowing for dynamic model adaptation and maintaining backward compatibility.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If AI models are used in encoding and decoding processes, then coding performance is improved, but the static nature of standardized AI models hinders long-term evolution and performance optimization
Solution Approach 1:
The patent introduces a dynamics principle by enabling the AI models to be updated and switched over time. The encoder and decoder can synchronously update their AI models through receiving model information in the bitstream, transforming the static standardized models into dynamic, evolvable systems that can adapt to new requirements while maintaining backward compatibility through version management.
2Adaptability or versatility
If model information is transmitted in the bitstream, then model updates and switches are enabled, but bit rate overheads and bandwidth requirements increase
Solution Approach 1:
The patent extracts only the essential model information needed for updates from the complete AI model, transmitting only the differential or critical parameters in the bitstream rather than the entire model. This selective extraction reduces the bit rate overhead while still enabling effective model updates and switches at the decoder side.
3Measurement precision
If encoding models are updated frequently, then coding performance optimization is achieved, but synchronization between encoder and decoder becomes more difficult
Solution Approach 1:
The patent implements a feedback mechanism where the encoder transmits model information and version identifiers in the bitstream to the decoder. The decoder uses this feedback to synchronously update its model to match the encoder's model, ensuring both sides operate with consistent AI models even during frequent updates, thereby managing synchronization complexity through structured information exchange.
Data Source
AI summary
A first electronic device obtains a to-be-encoded signal, and then generates a first bitstream, where the first bitstream includes a second bitstream and a model identifier. The second bitstream is obtained by encoding the to-be-encoded signal based on an encoding model, where a value of the model identifier is a preset identifier value, the preset identifier value indicates a second electronic device to obtain model information of a decoding model, and the decoding model corresponds to the encoding model. In this way, after the first bitstream is transmitted to the second electronic device, the second electronic device may obtain the model information of the decoding model based on the preset identifier value, to synchronously update the decoding model of the second electronic device and the encoding model of the first electronic device.