MMVD Candidate Segmentation and Filter Reordering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies, such as HEVC, face inefficiencies in resource usage when generating candidates for Merge Mode with Motion Vector Difference (MMVD), particularly due to large access ranges required for generating reference samples, which increase hardware costs and complexity.
Innovation Solution
The proposed video coding system reduces resource usage by dividing MMVD candidates into groups for separate processing stages, applying vertical filtering before horizontal filtering to generate reference samples, and using reduced filter sizes, thereby minimizing the access range and computational resources needed for MMVD mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the access range is enlarged to generate reference samples for all MMVD candidates, then the prediction accuracy is improved, but the hardware cost and device complexity increase
Solution Approach 1:
The patent divides MMVD candidates into two groups: first group candidates (vertical offset only) and second group candidates (horizontal offset only). This segmentation allows the system to generate reference samples for each group separately with reduced access ranges, thereby maintaining prediction accuracy while reducing hardware requirements and complexity.
2Measurement precision
If the access range is enlarged to generate reference samples for all MMVD candidates, then the prediction accuracy is improved, but the computational resources increase
Solution Approach 1:
By segmenting MMVD candidates into vertical-offset group and horizontal-offset group, the patent enables separate processing stages with reduced access ranges. This segmentation significantly reduces the computational burden for generating reference samples while maintaining effective prediction accuracy.
Solution Approach 2:
The patent applies vertical filtering before horizontal filtering in a sequential manner. This preliminary action (vertical filtering first) reduces the computational complexity compared to applying both filters simultaneously or in reverse order, as it prepares the data in an optimized sequence that reduces subsequent processing requirements.
3Use of energy by moving object
If vertical filtering is applied before horizontal filtering, then the computational resources are reduced, but the processing steps increase
Solution Approach 1:
The patent implements vertical filtering as a preliminary step before horizontal filtering. This sequential approach, while increasing processing steps, reduces the overall computational resource requirement by optimizing the filter application order. The vertical filtering prepares the data in a way that reduces the complexity of subsequent horizontal filtering operations.
Data Source
AI summary
A video coding system generating candidates for Merge Mode with Motion Vector Difference (MMVD) with reduced resource usage is provided. The system receives data to be encoded or decoded as a current block of a current picture of a video. The system identifies multiple MMVD candidates for different offset positions based on a merge candidate of the current block. The system generates reference samples for the identified MMVD candidates. The system reconstructs the current block or encodes the current block into a bitstream by using the generated reference samples. The system processes the MMVD candidates in separate groups: a first group of vertical MMVD candidates and a second group of horizontal MMVD candidates. The system generates the reference samples for the identified MMVD candidates by applying a vertical filter to source reference samples of horizontal MMVD candidates and then applying a horizontal filter to outputs of the vertical filter.