Optical Flow Motion Vector Refinement Without Extra Memory
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in accurately refining motion vectors without increasing memory costs, which affects the precision of video decoding and frame reconstruction.
Innovation Solution
The use of optical flow-based motion vector refinement techniques that incorporate extra samples and gradient values beyond the current subblock to derive refined motion vectors, maintaining the same block size for final motion compensation, thereby improving accuracy without additional memory costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector refinement uses only samples within the current subblock, then memory requirements are minimized, but motion vector accuracy deteriorates
Solution Approach 1:
The patent extends the optical flow refinement to operate in an overlapping region that spans multiple blocks, effectively adding a spatial dimension beyond the current subblock boundaries. This allows the refinement process to access gradient values and samples from neighboring blocks, improving motion vector accuracy without proportionally increasing memory requirements for the current subblock processing.
Solution Approach 2:
The patent performs preliminary computation of gradient values and optical flow parameters using samples from neighboring blocks before finalizing the motion vector refinement for the current subblock. This preliminary action enables the use of additional spatial information to improve accuracy while maintaining efficient memory usage during the actual refinement process.
2Measurement precision
If block size for optical flow refinement is increased to use more samples, then motion vector accuracy improves, but computational complexity increases
Solution Approach 1:
The patent segments the video data into blocks and subblocks, and applies optical flow refinement at the subblock level within a larger block context. This segmentation allows the refinement process to use samples from multiple blocks while maintaining manageable computational complexity through localized processing at the subblock level rather than processing the entire block as a single unit.
Solution Approach 2:
The patent applies different processing qualities to different regions: the current subblock receives refined motion vector processing using gradient values from overlapping regions, while maintaining the original block size for final motion compensation. This local quality approach improves accuracy where needed without unnecessarily increasing computational complexity across the entire block.
Data Source
AI summary
An example method of video coding includes receiving a video bitstream comprising a plurality of blocks. The method also includes deriving a set of subblock motion vectors for a current subblock in a first block of the plurality of blocks, and deriving a set of refined subblock motion vectors for the current subblock by applying an optical flow refinement that uses one or more values that are outside of the current subblock, where a block size for the optical flow refinement is a block size of the current subblock. The method further includes reconstructing the current subblock using the set of refined subblock motion vectors.