DMVR and BDOF Inter Prediction Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increasing demand for high-resolution and high-quality images and videos, particularly in immersive media formats like VR and AR, has led to a need for more efficient image and video compression technologies to reduce transmission and storage costs while maintaining quality.
Innovation Solution
The implementation of Decoder-side Motion Vector Refinement (DMVR) and Bi-directional optical flow (BDOF) techniques to enhance inter prediction methods, allowing for conditional application based on specific conditions to optimize motion compensation and improve coding efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution and high-quality image/video data are transmitted or stored using existing methods, then image quality is maintained, but transmission costs and storage costs are increased
Solution Approach 1:
The patent extracts and transmits only the essential visual information by performing inter prediction and generating residual data, which contains only the differences between original and predicted blocks. This extraction approach allows high-quality reconstruction at the decoder while significantly reducing the amount of data that needs to be transmitted or stored
Solution Approach 2:
The patent changes the parameter representation by transforming image blocks into frequency domain using transform processing, then applying quantization to modify the precision of transform coefficients. This parameter transformation enables efficient compression while preserving visually important information, reducing transmission costs while maintaining image quality
2Productivity
If DMVR and BDOF techniques are applied to enhance inter prediction, then coding efficiency is improved, but calculation complexity is increased
Solution Approach 1:
The patent applies DMVR and BDOF techniques selectively rather than universally. The encoder determines whether to apply these refinement techniques based on block characteristics, prediction mode, and other conditions. This partial application approach improves coding efficiency for suitable blocks while avoiding unnecessary calculation complexity for blocks where refinement provides minimal benefit
Solution Approach 2:
The patent divides the current block into multiple sub-blocks and applies motion compensation and refinement techniques independently to each sub-block. This segmentation allows the encoder to apply DMVR/BDOF only to specific sub-blocks that benefit from refinement, reducing overall calculation complexity while maintaining coding efficiency where needed
3Measurement precision
If motion vector refinement is performed for all blocks, then prediction accuracy is improved, but processing time is increased
Solution Approach 1:
The patent performs motion vector refinement only for blocks that meet specific criteria, such as certain prediction modes, block sizes, or motion characteristics. By applying refinement selectively rather than to all blocks, the system achieves improved prediction accuracy for critical blocks while avoiding the processing time penalty of universal refinement
Solution Approach 2:
The patent applies different levels of motion compensation quality to different blocks based on their local characteristics. Blocks with significant motion or important visual content receive refined motion vectors through DMVR/BDOF, while blocks with minimal motion use standard motion compensation. This local quality approach optimizes the balance between prediction accuracy and processing time
Data Source
AI summary
A video decoding method includes: deriving an LO motion vector and an LI motion vector for a current block; deriving a refined LO motion vector and a refined L1 motion vector, by applying DMVR to the current block based on DMVR flag information; deriving LO prediction samples of the current block based on the refined LO motion vector and LI prediction samples of the current block based on the refined L1 motion vector; deriving prediction samples for the current block based on the LO prediction samples and the L1 prediction samples, by applying BDOF to the current block based on BDOF flag information; and generating reconstruction samples for the current block based on the prediction samples, in which the DMVR flag information is derived based on a predetermined DMVR application condition, and the BDOF flag information is derived based on a predetermined BDOF application condition.