Reduced-Precision Multiplier for Video Coding Parameter Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing video coding systems face challenges in optimizing parameters for affine CPMV refinement and ALF refinement due to large data range requirements, which necessitate high precision floating-point operations, increasing computational complexity and resource demands.
Innovation Solution
The implementation of reduced-precision floating-point multiplication and reciprocal operations, using a truncated mantissa part and lookup tables, to derive updated parameters for affine CPMV and ALF refinements, reducing the bit depth requirements and computational complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high precision floating-point operations are used for affine CPMV refinement and ALF refinement, then parameter optimization accuracy is improved, but computational complexity and resource requirements increase
Solution Approach 1:
The patent changes the precision parameter of floating-point operations from high precision to low precision (e.g., using 5-bit or 6-bit mantissa instead of standard 24-bit or 53-bit). This parameter change enables the system to achieve sufficient parameter optimization accuracy while dramatically reducing computational complexity and resource requirements, directly resolving the technical contradiction between precision and complexity.
Solution Approach 2:
The patent applies partial precision to floating-point operations, using only the necessary bits (5-6 bits) for the mantissa rather than full precision. This partial action approach provides sufficient accuracy for video coding refinements while avoiding the excessive computational burden of high precision operations, effectively balancing accuracy and complexity.
2Manufacturing precision
If high precision floating-point operations are used for statistical data processing, then optimization results are improved, but encoding time increases
Solution Approach 1:
The patent changes the precision parameter of floating-point operations to use lower bit depths (5-6 bits for mantissa). This parameter change reduces the number of computational operations required during statistical data processing for affine CPMV and ALF refinements, thereby decreasing encoding time while maintaining sufficient optimization results quality.
Solution Approach 2:
The patent substitutes high-precision floating-point arithmetic with low-precision arithmetic operations. This substitution maintains the functional capability of statistical data processing and optimization while significantly reducing the computational time and processing resources required during encoding.
3Device complexity
If reduced-precision floating-point operations are used, then computational complexity is reduced, but data range coverage may be limited
Solution Approach 1:
The patent changes the precision parameter to use lower bit depths (5-6 bits for mantissa) while compensating for reduced data range coverage through careful selection of operating ranges and statistical processing techniques. This parameter change achieves computational complexity reduction while maintaining sufficient data range coverage for video coding applications through adaptive processing.
Data Source
AI summary
Method and apparatus for affine CPMV or ALF refinement are mentioned. According to this method, statistical data associated with the affine CPMV or ALF refinement are collected over a picture area. Updated parameters for the affine CPMV refinement or the ALF refinement are then derived based on the statistical data, where a process to derive the updated parameters includes performing multiplication using a reduced-precision multiplier for the statistical data. The reduced-precision multiplier truncates at least one bit of the mantissa part. In another embodiment, the process to derive the updated parameters includes performing reciprocal for the statistical data using a lookup table with (m−k)-bit input by truncating k bits from the m-bit mantissa part, and contents of the lookup table includes m-bit outputs. m and k are positive integers.