Gradient-Based Prediction Refinement Unified Displacement Precision
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding technologies face inefficiencies due to the need for different logic circuitry to handle varying precision levels for horizontal and vertical displacements in different inter-prediction modes, leading to increased size and power consumption.
Innovation Solution
The implementation of gradient-based prediction refinement, where the precision level of horizontal and vertical displacements is unified across different inter-prediction modes, allowing the same logic circuitry to be used for all modes by rounding displacements to a single precision level.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If different precision levels are used for horizontal and vertical displacements in different inter-prediction modes, then prediction accuracy is improved, but device complexity and power consumption increase due to the need for different logic circuitry
Solution Approach 1:
The patent applies universality by designing a single logic circuitry that can handle multiple precision levels (1/16, 1/64, 1/256) through configurable parameters. The same circuit performs gradient-based prediction refinement for different inter-prediction modes (AFFINE, BDOF) by adjusting precision control signals, eliminating the need for separate dedicated circuits for each mode and precision level.
Solution Approach 2:
The patent implements dynamics by making the precision level configurable and adjustable based on the specific inter-prediction mode being used. The system dynamically selects appropriate precision levels (e.g., 1/16 for AFFINE mode, 1/64 or 1/256 for BDOF mode) through control signals, allowing the logic circuitry to adapt its operation to match the requirements of each mode while maintaining a unified hardware structure.
2Measurement precision
If different precision levels are used for horizontal and vertical displacements in different inter-prediction modes, then prediction accuracy is improved, but power consumption increases due to multiple logic circuitry units
Solution Approach 1:
The patent applies universality by designing a single logic circuitry that can handle multiple precision levels (1/16, 1/64, 1/256) through configurable parameters. The same circuit performs gradient-based prediction refinement for different inter-prediction modes (AFFINE, BDOF) by adjusting precision control signals, eliminating the need for separate dedicated circuits for each mode and precision level.
Solution Approach 2:
The patent applies discarding and recovering by selectively activating only the precision level required for the current inter-prediction mode. When AFFINE mode is used, only 1/16 precision circuitry is activated; when BDOF mode is used, only 1/64 or 1/256 precision circuitry is activated. This selective activation reduces power consumption by keeping unused precision circuitry in a low-power state.
3Measurement precision
If multiple logic circuitry units are implemented to support different precision levels, then prediction accuracy for different modes is improved, but manufacturing cost and device size increase
Solution Approach 1:
The patent applies universality by designing a single logic circuitry that can handle multiple precision levels (1/16, 1/64, 1/256) through configurable parameters. The same circuit performs gradient-based prediction refinement for different inter-prediction modes (AFFINE, BDOF) by adjusting precision control signals, eliminating the need for separate dedicated circuits for each mode and precision level.
Solution Approach 2:
The patent applies parameter changes by modifying the precision control parameters of the unified logic circuitry based on the selected inter-prediction mode. Instead of changing the physical circuit structure, the system changes operational parameters (precision levels, shift amounts) to adapt to different modes, simplifying manufacturing and reducing device complexity.
Data Source
AI summary
This disclosure describes gradient-based prediction refinement. A video coder (e.g., video encoder or video decoder) determines one or more prediction blocks for inter-predicting a current block (e.g., based on one or more motion vectors for the current block). In gradient-based prediction refinement, the video coder modifies one or more samples of the prediction block based on various factors such as displacement in a horizontal direction, the horizontal gradient, a displacement in the vertical direction, and a vertical gradient. This disclosure provides for gradient-based prediction refinement where a precision level of the displacement (e.g., at least one of the horizontal or vertical displacement) is unified (e.g., the same) for different prediction modes.