Geometric Partitioning Mode with Motion Vector Differences for Video Coding

The patent divides the video block into multiple geometric partitions (first geometric partition and second geometric partition) with different partition ratios. Each partition is processed separately with its own motion information, allowing more flexible and efficient motion modeling compared to treating the entire block uniformly, thus improving coding efficiency.

The patent applies different motion modeling techniques to different geometric partitions based on their specific characteristics. By determining motion information independently for each partition and using partition-specific parameters (such as different motion vector difference handling), the system achieves optimized coding efficiency for each local region rather than using a one-size-fits-all approach.

The patent combines motion information from multiple geometric partitions into a unified representation that can be efficiently signaled in the bitstream. By merging the motion modeling approach across partitions while maintaining partition-specific accuracy, the system achieves precise motion conveyance without proportionally increasing bitstream complexity.

The patent applies motion vector differences selectively to geometric partitions where needed, rather than uniformly applying complex motion modeling to all partitions. This partial application strategy maintains motion information accuracy for critical regions while reducing overall bitstream complexity by using simpler modeling where appropriate.