Adaptive Bi-directional Optical Flow for Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding techniques, such as those using bi-directional optical flow (BDOF), often require multiple passes to refine motion vectors, which increases computational complexity without always improving prediction quality.
Innovation Solution
A method where a video coder determines whether to apply multi-iterative BDOF or one-pass BDOF to the initial motion vector based on the size of the current block, thereby reducing computational complexity without significantly affecting coding quality.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple passes of BDOF are applied to refine motion vectors, then prediction quality is improved, but computational complexity increases
Solution Approach 1:
The patent changes the parameter of BDOF passes from fixed multiple passes to adaptive one-pass based on block size. For blocks of 64x64 samples or larger, multi-pass BDOF is applied to improve prediction quality, while for smaller blocks, one-pass BDOF is used to reduce computational complexity. This parameter adaptation resolves the contradiction by tailoring the processing depth to the specific characteristics of each block.
Solution Approach 2:
The patent segments the video picture into different block sizes and applies different BDOF processing strategies to each segment. By dividing the processing based on block dimensions (64x64 samples threshold), the system can optimize between prediction quality and computational complexity for different regions of the video data independently.
2Measurement precision
If multi-iterative BDOF is applied to all blocks, then prediction quality is improved, but processing time increases
Solution Approach 1:
The patent adapts the number of BDOF passes as a parameter based on block size. By changing from a fixed multi-pass approach to a variable pass count (one-pass for small blocks, multi-pass for large blocks), the system reduces processing time for blocks where multiple passes provide minimal benefit while maintaining quality where it matters most.
Solution Approach 2:
The patent applies partial action by using one-pass BDOF for blocks where multi-pass would be excessive and provide minimal improvement. This partial application of the full BDOF process reduces overall processing time while still providing sufficient prediction quality for the majority of blocks.
3Measurement precision
If multi-pass BDOF is applied uniformly, then coding quality is improved, but energy consumption increases
Solution Approach 1:
The patent changes the energy consumption parameter by adapting the number of BDOF passes to block size. Smaller blocks use one-pass processing which consumes less energy, while only large blocks (64x64 samples or larger) receive multi-pass processing. This parameter adaptation reduces overall energy consumption while maintaining coding quality where it is most beneficial.
Solution Approach 2:
The patent segments the processing workload based on block dimensions, applying different energy consumption profiles to different segments. This segmentation allows the system to optimize the balance between coding quality and energy consumption by treating different block sizes differently rather than applying uniform processing.
Data Source
AI summary
A video decoder may be configured to derive an initial motion vector for a current block of video data; determine whether to apply multi-iterative bi-directional optical flow (BDOF) or one-pass BDOF to the initial motion vector based on a size of the current block; in response to determining to apply the one-pass BDOF to the initial motion vector, determine a modified motion vector for the current block using the one-pass BDOF; determine a prediction block, for the current block, corresponding to the modified motion vector; and use the prediction block to determine a decoded version of the current block.