Neural Network Filter Parameter Update in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional video coding techniques lack the ability to dynamically update filter parameters during the coding process, limiting the adaptability and efficiency of neural network-based filtering in video coding.
Innovation Solution
The proposed solution involves allowing filter parameters of a neural network filter model to be updated during the coding process, with indications of parameter changes included in the bitstream, enabling the use of different parameter sets for different video units and updating parameters based on coded information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If filter parameters are fixed for a neural network filter model, then the model structure remains simple and stable, but the adaptability to different video content characteristics is limited
Solution Approach 1:
The patent implements dynamic filter parameters that can be updated during the coding process. The filter model receives parameter updates from the bitstream, allowing the same neural network structure to adapt to different video content characteristics without changing the model architecture itself. This resolves the contradiction by making the filter parameters dynamic while keeping the model structure stable.
Solution Approach 2:
The patent changes the state of filter parameters from fixed to variable by introducing parameter update mechanisms in the bitstream. Different parameter sets can be selected and applied based on video content characteristics, enabling the filter to adapt to various scenarios while maintaining a single reusable model structure.
2Manufacturing precision
If a single neural network filter model is used for all video units, then the coding process is efficient and simple, but the processing quality for diverse video content is limited
Solution Approach 1:
The patent applies local quality by allowing different filter parameters to be applied to different video units or regions based on their specific characteristics. Each video unit can receive customized parameter settings from the bitstream, enabling high-quality filtering adapted to local content requirements while still using a single underlying model structure.
Solution Approach 2:
The patent makes a single neural network filter model universal by enabling it to process different video content types through parameter updates. The same model structure can be reused across various video units with different characteristics by loading appropriate parameter sets, achieving both high filtering quality and coding efficiency.
3Adaptability or versatility
If filter parameters are updated frequently during coding, then the adaptability to video content improves, but the computational overhead increases
Solution Approach 1:
The patent implements partial parameter updates where only necessary filter parameters are updated during the coding process based on actual video content needs. Rather than updating all parameters frequently, the system selectively updates only those parameters that require adaptation, reducing computational overhead while maintaining necessary adaptability.
Data Source
AI summary
A method of processing video data including determining, for a conversion between a video and a bitstream of the video, that the bitstream includes an indicator. The indicator indicates that a first parameter set for a neural network (NN) filter model includes different filter parameters than a second parameter set for the NN filter model. The method further includes performing the conversion based on the indicator. A corresponding video coding apparatus and non-transitory computer readable medium are also disclosed.