Bit-width Control for Bi-directional Optical Flow in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing bi-directional optical flow (BDOF) processing methods for video coding face challenges in optimizing bit-width control for intermediate parameters, leading to suboptimal precision in motion refinements and inefficient utilization of dynamic range in hardware implementation.
Innovation Solution
The proposed method introduces additional right shifts to lower the internal bit-width of gradient values and adjusts bit-shifts in the calculation of variables ψx(i,j), ψy(i,j), and θ(i,j) to control the entire BDOF process, ensuring appropriate internal bit-widths and preventing unnecessary precision loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If additional right shifts are introduced to lower internal bit-width of gradient values, then hardware implementation efficiency is improved, but precision of motion refinement may be degraded
Solution Approach 1:
The patent changes the bit-width parameters of intermediate calculation results by introducing additional right shifts. Specifically, it lowers the internal bit-width of gradient values (ψx(i,j) and ψy(i,j)) and adjusts bit-shifts in the calculation of θ(i,j) to optimize hardware implementation while maintaining sufficient precision through careful parameter selection.
Solution Approach 2:
The patent applies different bit-width control strategies to different intermediate parameters in the BDOF process. It selectively controls the bit-width of gradient values, summation results, and difference calculations with tailored right shift amounts, ensuring each parameter uses appropriate precision for its specific role in the computation.
2Device complexity
If bit-width of intermediate parameters is reduced, then computational complexity is lowered, but dynamic range utilization in hardware is suboptimal
Solution Approach 1:
The patent dynamically adjusts bit-shift amounts based on the specific calculation stage and parameter being processed. It introduces additional right shifts at different stages (gradient calculation, summation, difference calculation) with varying magnitudes, allowing the system to adapt precision to computational needs while maintaining full utilization of 32-bit hardware dynamic range.
3Manufacturing precision
If additional right shifts are applied to gradient values and summation, then bit-width control is improved, but precision loss may occur without proper adjustment
Solution Approach 1:
The patent incorporates feedback mechanisms by carefully analyzing the impact of each right shift operation on subsequent calculations. It adjusts bit-shift amounts based on the specific computational context, ensuring that precision is maintained where critical while allowing reduction where acceptable, thereby preventing unnecessary information loss.
Data Source
AI summary
A method for controlling bit-width for bi-directional optical flow (BDOF) for video coding includes encoding a first reference picture and a second reference picture, the second reference picture being different than the first reference picture, wherein the first reference picture is displayed before a current picture and the second reference picture is displayed after the current picture. A motion refinement of a coding unit (CU) is calculated by minimizing a difference between a first prediction L0 and a second prediction L1. First gradient values for the first prediction L0 and second gradient values for the second prediction L1 are calculated. A final bi-prediction of the CU is calculated. Also disclosed are an associated apparatus and an associated non-transitory computer readable storage medium.