Bi-Directional Optical Flow Bit-Width Control for 32-Bit Decoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing bi-directional optical flow (BDOF) designs in video coding, such as in Versatile Video Coding (VVC), suffer from suboptimal bit-width control of intermediate parameters, leading to unnecessary precision loss and inefficiency in motion refinement due to high-pass gradient calculations and noise sensitivity.

Innovation Solution

A method for controlling bit-widths of intermediate parameters in BDOF by introducing additional right shifts and clipping operations to optimize gradient calculations and motion refinement, ensuring the process operates within a 32-bit hardware implementation without exceeding 32 bits.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If high-pass gradient calculations are used in BDOF, then motion refinement precision is improved, but noise sensitivity increases and precision is lost due to suboptimal bit-width control

Engineering Contradiction:
Improvemotion refinement precisionVSAvoidnoise sensitivity
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent changes the bit-width parameters of intermediate calculations in the BDOF process. Specifically, it controls the bit-width of gradient calculations and motion refinement intermediate values to prevent precision loss while avoiding noise amplification. The method adjusts parameter precision dynamically throughout the calculation chain to maintain optimal signal-to-noise ratio.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If additional right shifts and clipping operations are introduced, then bit-width control is optimized and precision is maintained, but computational complexity increases

Engineering Contradiction:
Improveintermediate parameter precisionVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies right shifts and clipping operations at predetermined stages in the BDOF calculation process. These operations are performed in advance on intermediate parameters (gradients, motion refinements) to control their bit-width before they propagate through subsequent calculations. This preliminary control prevents precision loss accumulation without requiring complex dynamic adjustment mechanisms.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If 32-bit hardware implementation is utilized, then processing efficiency is improved, but bit-width overflow may occur without proper control

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidbit-width overflow prevention
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent implements dynamic bit-width control throughout the BDOF calculation chain. Each intermediate parameter (horizontal gradients, vertical gradients, sample differences, motion refinements) has its bit-width adjusted according to its specific role in the calculation. This dynamic control ensures that no intermediate value exceeds 32 bits while maximizing the utilization of available hardware precision for optimal processing efficiency.

Inventive Principle:
Principle #15Dynamics

Data Source

PatentUS20250287014A1Methods and devices for bit-width control for bi-directional optical flow
Publication Date: 2025.09.11 BEIJING DAJIA INTERNET INFORMATION TECH CO LTD
  • US20250287014A1 patent drawing
  • US20250287014A1 patent drawing
  • US20250287014A1 patent drawing

AI summary

The present disclosure relates to a bit-width control method of bi-directional optical flow (BDOF) for decoding a video signal. The method includes obtaining a first reference picture and a second reference picture associated with a 4×4 video sub-block; obtaining first prediction samples I(0) for samples in the 4×4 video sub-block from the first reference picture; obtaining second prediction samples I(1) for the samples in the 4×4 video sub-block from the second reference picture; obtaining motion refinements for the samples in the 4×4 video sub-block based on a 6×6 window applied to the 4×4 video sub-block, by controlling bit-widths of intermediate parameters; and obtaining bi-prediction samples of the samples in the 4×4 video sub-block based on the motion refinements.