Template-Matched Motion Vector Resolution in Bi-Prediction and Affine Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face challenges in efficiently reducing redundancy and improving compression efficiency, particularly in intra and inter-picture prediction, due to the increasing complexity of motion vector prediction and intra prediction modes, which result in higher bit requirements and reduced compression ratios.
Innovation Solution
The implementation of template matching (TM) on motion vector resolution (MVR) pairs for bi-prediction and affine mode, where TM costs are determined for each MVR pair based on current and reference templates, allowing for the selection of the most efficient MVR pair for reconstructing the current block.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If template matching is performed on multiple motion vector resolution pairs to select the most efficient MVR pair, then compression efficiency is improved and bit allocation is optimized, but computational complexity and processing time increase
Solution Approach 1:
The patent applies partial action by performing template matching only on a limited set of candidate MVR pairs rather than exhaustively searching all possible resolutions. The encoder evaluates a subset of MVR pairs (e.g., a few candidate resolutions) and selects the best match, rather than computing template matching costs for all possible MVR combinations. This reduces computational complexity while still achieving improved compression efficiency through adaptive MVR selection.
2Measurement precision
If adaptive motion vector resolution is used with bi-prediction and affine mode, then prediction accuracy is improved, but bit requirements and processing complexity increase
Solution Approach 1:
The patent implements dynamics by making the motion vector resolution adaptive rather than fixed. The encoder dynamically selects the appropriate MVR from multiple candidate resolutions based on the specific block characteristics and prediction mode (bi-prediction or affine mode). This allows the system to adjust the level of precision and computational effort according to local content requirements, improving prediction accuracy for complex regions while maintaining efficiency for simpler regions.
Solution Approach 2:
The patent changes the parameter of motion vector resolution from a fixed value to a variable that can take multiple discrete values. By introducing multiple candidate MVRs (e.g., different precision levels such as 1/4-pel, 1/8-pel, 1/16-pel) and selecting the optimal one through template matching, the system adapts the resolution parameter to match the local complexity of the video content, thereby improving prediction accuracy without uniformly increasing complexity across all blocks.
3Loss of information
If multiple motion vector resolution candidates are evaluated through template matching, then bit allocation is optimized, but encoding time and processing load increase
Solution Approach 1:
The patent applies preliminary action by pre-defining a set of candidate motion vector resolutions before the actual encoding process. These candidate MVRs are prepared in advance (e.g., a fixed set of standard resolutions like 1/4-pel, 1/8-pel, 1/16-pel), and the encoder simply needs to evaluate them through template matching and select the best match. This pre-preparation reduces the computational burden during real-time encoding compared to dynamically generating and evaluating all possible resolutions on the fly.
Data Source
AI summary
Aspects of the disclosure provide a method and an apparatus including processing circuitry that performs template matching (TM) on motion vector resolution (MVR) pairs of a current block by determining a TM cost corresponding to each one of the MVR pairs. The current block is predicted with an affine adaptive motion vector prediction (ΔMVP) mode. Each MVR pair includes a first MVR in a first plurality of MVRs corresponding to a first reference picture and a second MVR in a second plurality of MVRs corresponding to a second reference picture. Each TM cost is determined based on a current template of the current block and a bi-predictor of the respective MVR pair. The bi-predictor is based on a first reference template in the first reference picture and a second reference template in the second reference picture. An MVR pair is selected based on the determined corresponding TM costs.