DMVR and BDOF Inter Prediction for Efficient Video Decoding

The patent applies parameter changes by dynamically adjusting prediction modes (DMVR and BDOF) based on picture type, block size, and motion characteristics. The encoder selects between different prediction algorithms and refinement levels according to specific parameters such as whether the picture is intra or inter predicted, block dimensions, and motion vector differences, thereby optimizing compression efficiency while controlling calculation complexity.

The patent implements partial action by applying DMVR and BDOF selectively rather than universally. The advanced prediction techniques are applied only to specific blocks or regions that benefit most from them, determined by checking conditions such as block size thresholds, motion vector precision requirements, and prediction mode compatibility. This selective application reduces overall calculation complexity while maintaining compression efficiency for critical areas.

The patent applies local quality by using different prediction and refinement strategies for different blocks based on their specific characteristics. Each block is evaluated individually against criteria such as size, motion complexity, and reference picture availability. Blocks requiring high precision receive DMVR or BDOF processing, while simpler blocks use standard prediction, thereby achieving high prediction performance where needed without uniformly increasing encoding complexity across the entire picture.

The patent segments the encoding process into multiple stages with different complexity levels. The encoding proceeds through initial prediction, condition checking, selective refinement (DMVR/BDOF application), and final output. This segmentation allows the encoder to process blocks in an organized manner, applying computational resources efficiently to only those segments (blocks) that require advanced processing, thus balancing prediction performance with encoding speed.