MMVD Prediction Unit Reduces Video Decoding Calculation Load

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

The existing method for MMVD prediction in video coding requires a large calculation amount due to the derivation of template matching costs for all MMVD difference vector candidates.

Innovation Solution

A video decoding apparatus that includes an MMVD prediction unit to derive a difference vector at a specific distance and direction, and a parameter decoder to decode an index indicating a difference vector from an MMVD candidate list, with the MMVD prediction unit performing a search by deriving a template matching cost for the difference vector and interpolating the MMVD candidate list based on the cost.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If template matching costs are derived for all MMVD difference vector candidates, then prediction accuracy is improved, but calculation amount increases

Engineering Contradiction:
Improveprediction accuracyVSAvoidcalculation amount
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent segments the MMVD difference vector candidates into multiple groups based on their characteristics (e.g., motion vector direction, distance). Instead of calculating template matching costs for all candidates uniformly, the method processes different groups with different strategies, reducing the overall calculation burden while maintaining prediction accuracy for the most promising candidates.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent changes the parameter of template matching cost calculation by applying it selectively rather than universally. It modifies the calculation approach by using simplified cost functions for certain candidate groups and full template matching only for selected candidates, thereby reducing the total calculation amount while preserving prediction quality.

Inventive Principle:
Principle #35Parameter changes

2Measurement precision

If all MMVD difference vector candidates are evaluated, then decoding precision is improved, but processing time increases

Engineering Contradiction:
Improvedecoding precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary actions by pre-sorting or pre-categorizing MMVD difference vector candidates based on their likelihood of being optimal. This preliminary organization allows the decoder to focus computational resources on the most promising candidates first, reducing the time required to achieve accurate decoding results.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements a skipping mechanism where less promising MMVD candidates are quickly evaluated or skipped using simplified criteria. By rushing through the evaluation of unlikely candidates with minimal computation and focusing detailed analysis only on top candidates, the overall processing time is reduced while maintaining decoding precision.

Inventive Principle:
Principle #21Skipping (Rushing through)

Data Source

PatentEP4546779A1Video decoding device
Publication Date: 2025.04.30 SHARP KK
  • EP4546779A1 patent drawingFigure 1
  • EP4546779A1 patent drawingFigure 2(a)~2(f)
  • EP4546779A1 patent drawingFigure 3

AI summary

Video coding and/or decoding apparatuses that can enhance coding efficiency are provided. A video decoding apparatus includes an MMVD prediction unit configured to obtain a motion vector by adding a difference vector at a prescribed distance and in a prescribed direction to a predicted motion vector of a target block, and a parameter decoder configured to decode an index for indicating a difference vector from an MMVD candidate list, wherein the MMVD prediction unit derives a difference vector at a specific distance and in a specific direction from the prescribed distance and the prescribed direction, performs a search by deriving a template matching cost for the difference vector, and derives the MMVD candidate list by interpolating a difference vector candidate according to the template matching cost, and in the search, only a subset of a first direction is searched at a distance included in a prescribed distance set, and only a subset of a second direction is searched at a distance included in a distance set other than the prescribed distance set.