Motion Vector Difference Sign Prediction in Video Decoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Signaling the sign of a motion vector difference (MVD) in video coding consumes significant bandwidth due to the need to convey polarity information, which is not efficiently handled in existing video coding standards.
Innovation Solution
The proposed solution involves predicting the sign of MVD coordinates for various coding modes, such as merge mode with MVD (MMVD), affine MMVD, geometric partitioning mode with MMVD, and multi-hypothesis prediction mode, by constructing motion vector candidates with possible sign values, sorting them based on cost, and determining the respective sign using a motion vector sign predictor index.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the sign of MVD coordinates is explicitly signaled in video coding, then the motion vector difference information is complete and accurate, but the bandwidth consumption increases significantly due to overhead signaling
Solution Approach 1:
The decoder determines the MVD sign by itself using a sign predictor index and cost-based evaluation of motion vector candidates, without requiring explicit sign signaling from the encoder. This self-determination mechanism eliminates the need for separate sign signaling while maintaining accurate MVD reconstruction.
Solution Approach 2:
A sign predictor index is introduced as an intermediary parameter that indirectly conveys sign information. Instead of directly signaling MVD signs, the encoder transmits a compact index that the decoder uses to select from pre-defined sign prediction candidates, significantly reducing the bit cost of sign transmission.
2Measurement precision
If motion vector candidates are constructed with all possible sign values and sorted by cost, then the most accurate MVD sign is selected, but the computational complexity increases
Solution Approach 1:
Instead of exhaustively searching all possible MVD sign combinations, the method uses a limited set of sign prediction candidates (typically 2-4 candidates) with predefined sign patterns. This partial action approach achieves sufficient accuracy while dramatically reducing the search space and computational complexity compared to exhaustive enumeration.
Solution Approach 2:
Motion vector candidates with different sign combinations are pre-constructed and sorted by cost before the actual decoding decision. By performing this preparation work in advance, the decoder can quickly select the optimal sign combination using a simple index lookup rather than performing complex real-time optimization.
Data Source
AI summary
A video decoder may be configured to construct motion vector candidates using possible sign values, respective magnitudes of motion vector difference components, and a motion vector predictor for a block of video data, wherein the possible sign values include a positive sign value and a negative sign value, sort the motion vector candidates based on a cost for each of the motion vector candidates to create a sorted list, determine a respective motion vector difference sign for each motion vector difference coordinate based on a motion vector sign predictor index and the sorted list, and decode the block of video data using the respective magnitudes of motion vector difference coordinates and the respective motion vector difference sign for each motion vector difference component.