EVC Decoding Complexity Metrics for Profile-Aware CU Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
EVC baseline and main profiles have different methods for picture partitioning, leading to inconsistencies in determining the number of samples and coding units, which complicates the decision-making process for software engineering institute (SEI) message complexity metrics.
Innovation Solution
Implementing EVC decoding complexity metrics by extracting information from a bitstream that includes a direct indication of the number of coding units with non-zero transform coefficients for each CU size, allowing for the prediction of decoding complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If different methods for picture partitioning are used in EVC baseline and main profiles, then profile-specific optimization is achieved, but inconsistency in determining the number of samples and coding units increases complexity
Solution Approach 1:
The patent segments the complexity metric determination by separating profile-specific parameters (maxCtBWidth, maxCtBHeight) from universal calculation logic. The SEI message structure divides complexity metrics into profile-dependent fields that can be independently configured for baseline and main profiles, while sharing common computational algorithms for determining coding unit counts and transform coefficients.
Solution Approach 2:
The patent employs parameter changes by introducing profile-specific maximum coding block width and height parameters (maxCtBWidth, maxCtBHeight) that modify the complexity metric calculation based on the active profile. These parameters allow the same base algorithm to adapt to different EVC profiles by changing numerical values rather than structural logic, reducing implementation complexity.
2Quantity of substance
If the length of syntax elements indicating the number of samples and coding units is decided by their largest value, then the SEI message size is minimized, but precision in representing smaller values is reduced
Solution Approach 1:
The patent implements dynamic syntax element length adjustment where the bit depth of complexity metric fields is determined by the actual maximum values in the bitstream rather than fixed maximums. The encoder calculates the required precision based on the specific profile parameters and content characteristics, then dynamically sets the syntax element length to match the actual data range, optimizing both size and precision adaptively.
Solution Approach 2:
The patent applies partial precision allocation by assigning different bit depths to different complexity metric fields based on their actual value ranges. Rather than using uniform maximum precision for all fields, the patent allocates sufficient precision only where needed (e.g., larger bit depth for transform coefficient counts in high-complexity profiles, smaller bit depth for coding unit counts in low-complexity scenarios), minimizing overall message size while maintaining necessary precision.
Data Source
AI summary
An apparatus includes a communication interface and a processor operably coupled to the communication interface. The processor is configured to receive, via the communication interface, a bitstream including complexity metrics (CM) metadata. The processor is also configured to extract information in the CM metadata from the bitstream, wherein the information includes a direct indication of a number of coding units (CUs) in the bitstream that have non-zero transform coefficients, for each CU size of a plurality of CU sizes. The processor is further configured to predict a decoding complexity of the bitstream based on information in the CM metadata.