Adaptive Motion Vector Resolution for Affine Video Coding

The patent implements dynamic motion vector resolution by allowing the resolution level to vary adaptively across different blocks and reference picture lists. The system selects from multiple resolution levels (e.g., 1/4 pel, 1/2 pel, 1 pel) based on local motion characteristics, enabling fine-grained motion areas to use higher resolution while coarser areas use lower resolution, thus improving coding efficiency without uniformly increasing complexity across the entire picture.

The patent applies different motion vector resolutions to different local regions (blocks) of the picture based on their specific motion characteristics. By using the amvr_precision_flag and amvr_precision_cb_flag at the block level, the system assigns appropriate resolution quality locally rather than uniformly, which improves compression for complex motion regions while maintaining acceptable quality for simpler regions, thereby balancing coding efficiency and complexity.

The patent segments the motion vector precision control into separate flags for different reference picture lists (amvr_precision_flag for list 0, amvr_precision_cb_flag for list 1). This segmentation allows independent optimization of precision for each reference list based on actual motion characteristics, avoiding the need to signal precision information for all possible cases, thus reducing overall bitstream overhead while maintaining high precision where needed.

The patent changes the precision parameter of motion vectors dynamically based on block characteristics and reference picture list. By using precision flags that indicate the actual resolution level (e.g., 0 for 1/4 pel, 1 for 1/2 pel, 2 for 1 pel), the system adapts the motion vector precision parameter to match local motion complexity, achieving high measurement precision for complex motions while using lower precision for simple motions, thereby reducing the average information loss.