Video Coding Motion Compensation With Uni-Prediction Limits

The patent applies local quality by differentiating the prediction mode based on prediction unit size. Small PUs (4×4, 8×4, 4×8) are restricted to uni-prediction while larger PUs can use bi-prediction. This localized approach optimizes memory bandwidth usage by applying bi-prediction only where it provides significant benefit (larger blocks), while accepting uni-prediction for smaller blocks where the memory overhead is less justifiable.

The patent changes the prediction mode parameter based on PU size. By introducing a size-based condition that switches between uni-prediction and bi-prediction modes, the system dynamically adjusts the prediction approach to balance accuracy requirements against memory bandwidth constraints. This parameter change resolves the contradiction by adapting the prediction strategy to the specific spatial scale of each prediction unit.

The patent applies local quality by restricting bi-prediction to specific PU sizes rather than uniformly applying it across all blocks. This selective approach ensures that the enhanced prediction accuracy of bi-prediction is used only where it provides the most benefit (larger PUs), while reducing memory bandwidth consumption for smaller PUs where the benefit is less significant.

The patent implements partial action by not enabling bi-prediction for all prediction units, but only for those meeting specific size criteria. This partial application of bi-prediction achieves a balance between coding efficiency improvement and memory bandwidth consumption, avoiding the excessive memory usage that would result from universal bi-prediction while still capturing the benefits where most needed.