Inter-Block Intra Prediction for Parallel Point Cloud Coding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional RAHT transform in point cloud compression cannot be parallelized, hindering real-time applications due to its requirement of processing the entire point cloud frame as a whole, preventing independent encoding of sub-blocks of the voxel grid.

Innovation Solution

Divide the point cloud into blocks for encoding/decoding, employing a global step to encode the average color of each block and a local step to encode attributes, allowing parallel execution of intra prediction using neighboring blocks for improved efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If conventional RAHT transform processes the entire point cloud frame as a whole, then encoding accuracy is maintained, but processing speed deteriorates and real-time applications are hindered

Engineering Contradiction:
Improveprocessing speedVSAvoidencoding complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent divides the point cloud frame into multiple independent sub-blocks (e.g., 8x8 or 16x16 voxel grids) that can be encoded and decoded in parallel. Each sub-block is processed independently with its own intra-prediction referencing neighboring sub-blocks, enabling parallelization while maintaining overall encoding quality through the global step that processes average colors at block boundaries.

Inventive Principle:
Principle #1Segmentation

2Productivity

If the entire point cloud frame is processed sequentially, then encoding accuracy is preserved, but real-time processing capability is lost

Engineering Contradiction:
Improvereal-time processing capabilityVSAvoidprocessing time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The point cloud attribute data is segmented into multiple independently encodable sub-blocks arranged in a grid pattern. Each sub-block can be processed simultaneously by different processing units, transforming sequential processing into parallel execution. The segmentation enables real-time processing by reducing the critical path length while maintaining compression efficiency through coordinated boundary handling.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The global step performs preliminary encoding of average color values for each block before the local step processes detailed attributes. This preliminary action establishes reference data at block boundaries that enables subsequent parallel local encoding without requiring global coordination during the actual attribute encoding phase, reducing processing time while preserving accuracy.

Inventive Principle:
Principle #10Preliminary action

3Productivity

If independent sub-block encoding is implemented, then parallel processing is enabled, but reference data availability across block boundaries becomes problematic

Engineering Contradiction:
Improveparallel processing capabilityVSAvoidreference data availability
Core Design Contradiction:
ProductivityVSLoss of information

Solution Approach 1:

The patent introduces a global step as an intermediary that processes average color values at block boundaries and generates reference data that is made available to all local steps. This intermediary layer ensures that neighboring blocks have access to necessary reference information for accurate intra-prediction, preventing information loss while enabling parallel processing of sub-blocks.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

Reference data for block boundaries is prepared in advance by the global step before the parallel local encoding begins. This preliminary preparation of reference information ensures that all sub-blocks have access to necessary data for accurate prediction without requiring inter-communication during parallel execution, maintaining information availability while enabling independence of parallel processes.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4633166A1Inter block multi-layer intra prediction for region-adaptive hierarchical transform
Publication Date: 2025.10.15 INTERDIGITAL CE PATENT HOLDINGS SAS
  • EP4633166A1 patent drawingFigure 1
  • EP4633166A1 patent drawingFigure 2
  • EP4633166A1 patent drawingFigure 3

AI summary

In a point cloud codec, attributes correspond to nodes in a tree structure. The tree is partitioned into subtrees, each subtree having a respective root node. The root nodes may all be at a layer Lglobal. Attributes of the subtree root nodes may be coded using intra coding, with residuals coded using a region-adaptive hierarchical transform (RAHT). To allow for parallel decoding, nodes of the subtrees may be intra predicted using neighboring nodes from different subtrees as reference nodes even if they have not yet been reconstructed, e.g. by temporarily using the attributes of the subtree root node as the attribute of one or more of the reference nodes for purposes of the intra prediction. Those temporary attributes may be replaced with the actual attributes once the nodes in the subtree have been reconstructed (e.g. using a local RAHT for each subtree).