View-Dependent Quantization for Geometric Data Compression

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing quantization techniques distribute quantized locations uniformly, leading to consistent quantization error throughout a scene, resulting in perceivable artifacts close to the viewpoint due to uniform coarse spacing of quantization points.

Innovation Solution

Implementing view-dependent quantization values that are less coarse close to the viewpoint and more coarse with increasing distance, using transformations like z'= 1/(1+z) to adjust distances and minimize error, allowing for adaptive quantization based on proximity to the viewpoint.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If uniform quantization distribution is used, then data representation consistency is maintained, but quantization error becomes perceivable as artifacts near the viewpoint

Engineering Contradiction:
Improvequantization precisionVSAvoidperceivable artifacts
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent applies local quality by making quantization precision view-dependent rather than uniform. Quantization values are adjusted based on distance from the viewpoint, with finer precision allocated to regions closer to the viewpoint and coarser precision to distant regions. This resolves the contradiction by maintaining high precision where it matters most (near the viewpoint) while accepting coarser precision in less critical areas, thereby eliminating perceivable artifacts without uniformly increasing data complexity throughout the entire scene.

Inventive Principle:
Principle #3Local quality

2Object-affected harmful factors

If view-dependent quantization is implemented, then artifacts near viewpoint are reduced, but quantization complexity increases

Engineering Contradiction:
Improveperceivable artifactsVSAvoidquantization complexity
Core Design Contradiction:
Object-affected harmful factorsVSDevice complexity

Solution Approach 1:

The patent implements parameter changes by transforming the quantization approach from uniform to view-dependent based on distance parameters. A transformation function (such as z' = 1/(1+z)) is applied to adjust quantization values according to their distance from the viewpoint. This resolves the contradiction by systematically varying quantization parameters based on spatial relationships, reducing artifacts through adaptive precision while maintaining manageable complexity through mathematical transformation rather than complex algorithms.

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If uniform quantization spacing is used, then implementation simplicity is maintained, but memory usage increases due to coarse spacing throughout the scene

Engineering Contradiction:
Improveimplementation simplicityVSAvoidmemory usage
Core Design Contradiction:
Ease of manufactureVSQuantity of substance

Solution Approach 1:

The patent applies local quality by implementing non-uniform quantization spacing that adapts to viewing distance. Regions closer to the viewpoint use finer spacing to preserve detail, while distant regions use coarser spacing to reduce data volume. This resolves the contradiction by optimizing memory usage through spatially-varying precision - storing more data only where it is visually important and less data where it is less critical, thereby reducing overall memory requirements while maintaining implementation feasibility through systematic distance-based categorization.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3603071B1Implicit view-dependent quantization
Publication Date: 2021.01.13 GOOGLE LLC
  • EP3603071B1 patent drawingFigure 1A
  • EP3603071B1 patent drawingFigure 1B
  • EP3603071B1 patent drawingFigure 2

AI summary

An encoder includes a processor and a memory. The memory includes code as instructions that cause the processor to receive geometric data and a viewpoint (POV) associated with the geometric data, and quantize the geometric data based on the viewpoint using a plurality of view-dependent quantization values (115) each having a quantized distance based on a proximity to the viewpoint.