Geometric Partition Mode Motion Vector Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as VVC and AVS3, face inefficiencies in the geometric partition mode (GPM) due to inaccurate motion estimation for video block partitions, leading to suboptimal coding performance and increased signaling overhead.
Innovation Solution
The proposed solution introduces motion vector refinement (MVR) for each geometric partition, allowing for more accurate motion estimation by applying predefined MVD magnitudes and directions, and signaling these refinements in a manner similar to the MMVD mode, enabling improved prediction samples generation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector refinement is applied to each geometric partition, then prediction accuracy is improved, but signaling overhead increases
Solution Approach 1:
The patent applies parameter changes by using predefined MVD magnitudes and directions instead of fully independent motion vectors for each partition. This restricts the parameter space that needs to be signaled, reducing overhead while maintaining refinement capability. The MVR parameters (magnitude and direction) are changed from full MV precision to coarser predefined levels.
Solution Approach 2:
The patent applies local quality by allowing different refinement strategies for different partitions. Each geometric partition can have its own MVR applied independently, enabling localized optimization where refinement is most beneficial while avoiding unnecessary refinement in areas where it provides minimal improvement.
2Measurement precision
If separate motion vectors are used for each geometric partition, then prediction accuracy is improved, but device complexity increases
Solution Approach 1:
The patent simplifies device complexity by changing the parameters from full motion vectors to motion vector refinements with predefined magnitudes and directions. This reduces the computational burden of selecting and processing multiple candidate vectors while maintaining the ability to apply separate refinements to each partition.
Solution Approach 2:
The patent applies segmentation by dividing the motion compensation process into two stages: base motion vector selection and refinement application. This segmentation allows the complex task of motion estimation to be broken down into manageable steps, reducing overall processing complexity.
3Productivity
If geometric partition mode is used, then coding efficiency is improved, but signaling overhead increases
Solution Approach 1:
The patent reduces bit rate overhead by changing the parameter representation from full motion vectors to compact MVR syntax elements. The predefined magnitude and direction indices require fewer bits to encode compared to transmitting complete motion vector values for each partition.
Solution Approach 2:
The patent uses copying by deriving motion information from a candidate list of existing motion vectors rather than encoding completely new vectors. This allows the system to copy and refine existing motion data, reducing the amount of new information that needs to be transmitted.
Data Source
AI summary
Methods, apparatuses, and non-transitory computer-readable storage mediums are provided for decoding a video block in geometry partition mode (GPM). The method may include partitioning the video block into first and second geometric partitions. The method may include obtaining a first motion vector refinement (MVR) for the first geometric partition. The method may include obtaining a second MVR for the second geometric partition. The method may include obtaining a candidate list that comprises a plurality of motion vectors (MVs). The method may include obtaining a first MV from the candidate list for the first geometric partition. The method may include obtaining a second MV from the candidate list for the second geometric partition. The method may include calculating a third MV by adding the first MV with the first MVR for the first geometric partition. The method may include calculating a fourth MV. The method may include obtaining prediction samples.