Distributed Decoding Refresh for Low-Delay Video Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video coding standards face challenges in achieving low delay and efficient bit-rate management, particularly with intra random access point (IRAP) pictures, which cause high bit-rate spikes and increased latency, while existing gradual decoding refresh (GDR) techniques suffer from coding efficiency penalties and complex implementation issues.

Innovation Solution

Distribute intra prediction pictures into sub-units across multiple other pictures, allowing random access and refresh without requiring IRAP pictures, using high-level syntax modifications to simplify implementation and reduce bit-rate variations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If intra random access point (IRAP) pictures are used for random access and refresh, then random access capability is improved, but bit-rate increases and latency increases

Engineering Contradiction:
Improverandom access capabilityVSAvoidbit-rate
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent segments the intra prediction picture into multiple sub-units (e.g., CTUs, CUs, or NAL units) and distributes them across multiple other pictures with the same POC value. This segmentation allows the refresh function to be spread out over time and multiple pictures, avoiding the need for a single large IRAP picture that causes bit-rate spikes.

Inventive Principle:
Principle #1Segmentation

2Reliability

If intra random access point (IRAP) pictures are used for random access and refresh, then random access capability is improved, but latency increases

Engineering Contradiction:
Improverandom access capabilityVSAvoidlatency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

By segmenting the intra prediction picture into sub-units distributed across multiple pictures, the patent enables incremental decoding and output. The decoder can start outputting pictures before receiving all sub-units of the intra prediction picture, thereby reducing latency compared to waiting for a complete IRAP picture.

Inventive Principle:
Principle #1Segmentation

3Quantity of substance

If gradual decoding refresh (GDR) techniques are used to reduce bit-rate, then bit-rate management is improved, but coding efficiency decreases and implementation complexity increases

Engineering Contradiction:
Improvebit-rate managementVSAvoidcoding efficiency
Core Design Contradiction:
Quantity of substanceVSProductivity

Solution Approach 1:

The patent applies segmentation by dividing the intra prediction picture into sub-units that are distributed across multiple pictures. This approach achieves smooth bit-rate management similar to GDR but with simpler implementation at the NAL unit level, avoiding the complex picture-level modifications and rate-control mechanisms required by traditional GDR techniques.

Inventive Principle:
Principle #1Segmentation

4Quantity of substance

If gradual decoding refresh (GDR) techniques are used to reduce bit-rate, then bit-rate management is improved, but implementation complexity increases

Engineering Contradiction:
Improvebit-rate managementVSAvoidimplementation complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent implements segmentation at the NAL unit level by dividing the intra prediction picture into sub-units and distributing them across multiple pictures. This high-level implementation approach simplifies the decoding process compared to traditional GDR, which requires complex picture-level modifications, region refresh information, and advanced rate-control mechanisms.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4618548A1Design for distributed decoding refresh in video coding
Publication Date: 2025.09.17 HUAWEI TECH CO LTD
  • EP4618548A1 patent drawingFigure 1A
  • EP4618548A1 patent drawingFigure 1B
  • EP4618548A1 patent drawingFigure 2

AI summary

It is provided a method of video encoding a sequence of pictures, the method being implemented in an encoding device and comprising: obtaining an intra prediction picture comprising at least two sub-units; obtaining a first encoded other picture, wherein the intra prediction picture and the first encoded other picture have the same Picture Order Count, POC, value; obtaining encoded other pictures different from the intra prediction picture and the first encoded other picture; and distributing the at least two sub-units of the intra prediction picture over one or more sub-units of the encoded other pictures and/or over one or more sub-units of the first encoded other picture.