Point Cloud Attribute Encoding With Fixed-Point Transforms
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing point cloud encoding methods using transformation matrices with floating point numbers lead to computational precision loss, affecting encoding accuracy.
Innovation Solution
Perform attribute transformation encoding using a transformation matrix that does not include floating point numbers, combined with layer-specific division and shift operations, and quantization, to generate a target bitstream.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a transformation matrix with floating point numbers is used for attribute transformation encoding, then the encoding process can be performed, but computational precision loss occurs affecting encoding accuracy
Solution Approach 1:
The patent changes the parameter type of the transformation matrix from floating point numbers to fixed point numbers (integers). This parameter change eliminates the precision loss inherent in floating point arithmetic while maintaining the functional capability of the transformation operation. The fixed point matrix operates on integer values throughout the encoding process, ensuring computational precision is preserved.
2Productivity
If a transformation matrix with floating point numbers is used, then the transformation can be performed, but computational complexity increases and precision is lost
Solution Approach 1:
The patent substitutes the floating point arithmetic system with a fixed point arithmetic system. This replacement eliminates the complex floating point unit requirements and associated computational overhead, simplifying the computational system while maintaining transformation functionality. The integer-based operations are computationally simpler and more efficient than floating point operations.
3Quantity of substance
If quantization processing is applied to attribute coefficients, then the data can be compressed, but precision may be reduced
Solution Approach 1:
The patent performs the transformation operation using fixed point arithmetic before quantization, ensuring that all intermediate calculations maintain integer precision. This preliminary use of exact arithmetic operations prevents precision loss that would otherwise occur during the transformation stage, preserving maximum precision through the quantization process while still enabling data compression.
Data Source
AI summary
An attribute transformation encoding method includes: generating a transform tree structure corresponding to a point cloud based on geometry information of the point cloud; performing a transformation operation on a first attribute coefficient corresponding to a child node of each first node in N layers by using a preset target transformation matrix, to determine a second attribute coefficient, and predicting a first attribute coefficient corresponding to each second node in the N layers, to determine an attribute coefficient residual; quantizing the second attribute coefficient, the attribute coefficient residual, and a first attribute coefficient corresponding to a child node of each first node in a top layer; and encoding the second attribute coefficient, the attribute coefficient residual, and the first attribute coefficient corresponding to the child node of each first node in the top layer of the N layers that are quantized and the geometry information, to generate a target bitstream.