Neural-Network Post-Filter Input Order for Video Bitstreams
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards lack efficient methods for utilizing neural-network post-filters (NNPFs) in video bitstreams, particularly in versatile video coding (VVC) and versatile supplemental enhancement information (VSEI) standards, leading to suboptimal video processing and increased bandwidth demand.
Innovation Solution
Implementing neural-network post-filters (NNPFs) with specified signaling and processing order in video bitstreams, allowing for the use of unfiltered or filtered pictures as inputs, and indicating whether NNPFs can be applied during decoding, enhancing video processing efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If neural-network post-filters are applied to improve video processing quality, then video processing efficiency is improved, but bandwidth demand increases
Solution Approach 1:
The patent applies preliminary action by determining the input picture status before NNPF application and pre-planning the processing sequence. The encoder/decoder determines whether the current picture or reference picture should be used as input to the NNPF, and establishes the processing order in advance, allowing optimal filtering without requiring additional bandwidth for multiple picture versions.
2Productivity
If multiple pictures are used as NNPF input to improve filtering quality, then video processing efficiency is improved, but processing complexity increases
Solution Approach 1:
The patent applies dynamics by making the NNPF input selection adaptive rather than fixed. The system dynamically determines whether to use the current picture or reference picture as NNPF input based on picture type (I-picture, P-picture, B-picture) and filtering conditions. This dynamic approach optimizes processing efficiency while managing complexity through context-based decision rules.
Solution Approach 2:
The patent applies parameter changes by modifying the input picture selection parameters based on picture type and filtering conditions. Different parameters are used for different picture types: I-pictures use current picture input, while P and B pictures may use reference picture input. This parameter adaptation allows efficient processing tailored to each picture's characteristics without uniform complexity.
Data Source
AI summary
A mechanism for processing video data is disclosed. The mechanism includes determining when an input of a neural-network post-filter (NNPF) activated for a current picture includes the current picture, the current picture used in the input is the current picture before applying the NNPF. A conversion is performed between a visual media data and a bitstream based on the NNPF.