Quality-Adaptive Neural In-Loop Filtering for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in effectively reducing distortion during compression, particularly in the context of bandwidth demand and quality considerations, which are not adequately addressed by conventional in-loop filtering methods.
Innovation Solution
Implementing neural network (NN) filter models for in-loop filtering, where the selection of these models is based on the reconstructed quality level of a video unit, and incorporating them into the bitstream with supplemental enhancement information (SEI) messages, allowing for adaptive filtering strategies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional in-loop filtering methods are used, then the filtering process is simple and fast, but the distortion reduction effectiveness is insufficient
Solution Approach 1:
The patent replaces conventional mechanical filtering methods (deblocking filter, SAO, ALF) with a neural network-based filtering system. The neural network model processes reconstructed video data to reduce distortion more effectively than traditional methods, achieving better visual quality while managing complexity through automated learning-based approaches
Solution Approach 2:
The patent introduces a quality level parameter that selects different neural network filter models based on the reconstructed quality of video data. This allows the system to adapt the filtering strength and complexity dynamically, using stronger filtering for lower quality levels and weaker filtering for higher quality levels, thus balancing distortion reduction with processing complexity
2Adaptability or versatility
If a single fixed filter model is used for all video data, then the system is simple to implement, but it cannot adapt to different reconstructed quality levels
Solution Approach 1:
The patent implements a dynamic filter selection mechanism where the system automatically chooses the appropriate neural network filter model based on the reconstructed quality level of the video data. This dynamic adaptation allows the system to optimize filtering performance for different quality conditions without manual intervention, balancing adaptability with implementation simplicity through automated quality-based selection
Solution Approach 2:
The patent segments the filter models into different categories based on quality levels (e.g., first quality level, second quality level, third quality level). Each segment corresponds to a specific neural network filter model designed for particular quality ranges, allowing the system to divide the complex filtering task into manageable segments and select the appropriate segment based on input quality
3Reliability
If multiple neural network filter models are implemented for different quality levels, then the distortion reduction performance improves, but the system complexity increases
Solution Approach 1:
The patent uses quality level parameters to select from multiple pre-trained neural network filter models. By organizing models according to quality levels (first, second, third quality levels with different QP ranges), the system can efficiently switch between models based on the reconstructed video quality, improving compression efficiency while managing complexity through parameter-based model selection rather than treating all models equally
Data Source
AI summary
A method of processing video data. The method includes selecting an in-loop filter from a plurality of neural network (NN) filter model candidates, wherein the plurality of NN filter model candidates are based on a reconstructed quality level of a video unit, and performing a conversion between a video media file comprising the video unit and a bitstream based on the in-loop filter selected. A corresponding video coding apparatus and non-transitory computer readable medium are also disclosed.