Neural Network Post-Filter Signaling for Video Padding Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding standards lack efficient methods for incorporating neural network post-filter techniques to enhance video quality, particularly in the context of emerging standards like VVC, where CNN-based post-filtering was not initially included.
Innovation Solution
The method involves signaling neural network post-filter characteristics through syntax elements, specifying padding types and sample values, enabling the application of neural network post-filters to reconstructed video data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing video coding standards are used without neural network post-filter techniques, then the coding process is simple and compatible with current standards, but video quality enhancement is limited
Solution Approach 1:
The patent introduces neural network post-filter characteristics messages as an intermediary component between the existing video coding standard and the desired quality enhancement. These messages carry padding type information and sample value specifications that enable CNN-based filtering without requiring modifications to the core coding standard, thus improving video quality while maintaining compatibility with current coding processes
Solution Approach 2:
The patent applies parameter changes by introducing new syntax elements that specify padding types (e.g., zero-padding, replication-padding) and sample values for the neural network post-filter. By changing and specifying these parameters through signaling mechanisms, the system enables flexible quality enhancement while keeping the base coding standard unchanged
2Manufacturing precision
If neural network post-filter techniques are incorporated into video coding standards, then video quality is enhanced, but the standards and coding process become more complex
Solution Approach 1:
The patent segments the neural network post-filter functionality into separate, independently specifiable parameters including padding type syntax elements and sample value syntax elements. This segmentation allows the complex neural network functionality to be broken down into manageable components that can be signaled and processed separately, reducing the apparent complexity in the standard while enabling sophisticated quality enhancement
3Measurement precision
If padding types and sample values are specified through syntax elements, then the neural network post-filter can be applied accurately, but the bitstream size increases
Solution Approach 1:
The patent implements partial signaling by introducing syntax elements for padding types and sample values that are conditionally included in the bitstream. Rather than signaling all possible parameters unconditionally, the system only signals the necessary padding and sample value information when neural network post-filter characteristics messages are present, achieving accurate filter application while minimizing bitstream overhead
Data Source
AI summary
A method of signaling neural network post-loop filter information for video data is disclosed. The method comprising: signaling a neural network post-filter characteristics message; signaling a syntax element specifying a padding type corresponding to the neural network post-filter characteristics message, wherein the syntax element includes a value specifying a wrap-around padding type and a value specifying a fixed padding type; and signaling a syntax element indicating a sample value to be used for padding in a case where the fixed padding type is specified.