RAHT Learning for Efficient Point Cloud Attribute Encoding

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for compressing and rendering point clouds in applications like autonomous driving, cultural heritage archival, and virtual/augmented reality are inefficient and lack effective techniques for predicting and encoding attributes of 3D points in point clouds.

Innovation Solution

A method involving a region-adaptive hierarchical transform (RAHT) learning-based process is used to predict attributes of child nodes in point clouds, utilizing a deep network model and multi-layer perceptron (MLP) for encoding, with adaptive weight determination through training and runtime processes.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional point cloud compression methods are used, then the compression process is simpler, but the compression efficiency and attribute prediction accuracy are insufficient

Engineering Contradiction:
Improvecompression efficiencyVSAvoidencoding complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent performs preliminary training to determine optimal weights for prediction data nodes before the actual encoding process. These pre-computed weights are stored and reused during encoding, eliminating the need to recalculate them for each point cloud, thus improving compression efficiency without significantly increasing encoding complexity

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent dynamically adapts the prediction process by determining weights based on the specific characteristics of each point cloud data set. The system can adjust which prediction data nodes are used and what weights are applied, making the compression process adaptive rather than static, thereby improving efficiency for different types of 3D data

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If more prediction data nodes and weights are used, then the attribute prediction accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improveattribute prediction accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies different weights to different prediction data nodes based on their local relevance to the target node. Instead of using uniform weights, the system determines specific weights for each prediction data node (such as parent, sibling, or neighbor nodes) based on their spatial and attribute relationships, improving prediction accuracy while keeping computational complexity manageable through localized processing

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent changes the parameters of the prediction process by dynamically determining weights based on point cloud characteristics. The system can adjust the number of prediction data nodes, their selection criteria, and their associated weights, allowing optimization of the balance between prediction accuracy and computational complexity for different applications

Inventive Principle:
Principle #35Parameter changes

3Adaptability or versatility

If learning-based processes are performed at runtime, then the adaptation to specific point clouds improves, but the processing time increases

Engineering Contradiction:
Improvepoint cloud adaptationVSAvoidprocessing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent performs the learning-based weight determination process during a preliminary training phase before actual encoding. The trained weights are then stored and applied during runtime without requiring additional learning computations, achieving both adaptation to specific point cloud characteristics and efficient real-time processing

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent implements a dynamic approach where the system can switch between pre-trained weights and runtime retraining based on requirements. For applications requiring high speed, pre-trained weights are used; for applications requiring maximum adaptation, runtime learning can be performed, providing flexibility in the adaptability-speed tradeoff

Inventive Principle:
Principle #15Dynamics

Data Source

PatentEP4636611A1RAHT learning based prediction
Publication Date: 2025.10.22 INTERDIGITAL CE PATENT HOLDINGS SAS
  • EP4636611A1 patent drawingFigure 1A
  • EP4636611A1 patent drawingFigure 1B
  • EP4636611A1 patent drawingFigure 1C

AI summary

Some embodiments of a method may include obtaining information corresponding to a point cloud, wherein the point cloud comprises a plurality of data nodes; selecting a parent node from the plurality of data nodes; determining a set of prediction data nodes within a neighborhood of the parent node; determining a set of weights corresponding to the set of prediction data nodes; predicting attributes corresponding to a child node of the parent node, wherein predicting the attributes uses the set of prediction data nodes and the set of weights corresponding to the set of prediction data nodes; encoding point cloud data using the predicted attributes, wherein the point cloud data corresponds to the plurality of data nodes; and transmitting the encoded point cloud data as a bitstream to a device.