Point Cloud Decoding Using Neural Network Residual Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing geometry-based point cloud compression (G-PCC) coding frameworks suffer from severe distortion and quality degradation due to lossy attribute encoding, resulting in a large difference between the reconstructed and original point clouds.
Innovation Solution
A decoding method that uses a preset network model to process geometry information and attribute values of a reconstructed point cloud, enhancing attribute quality and reducing distortion by learning a residual between the reconstructed and original point clouds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of substance
If lossy attribute encoding is used in G-PCC coding framework, then compression ratio is improved, but distortion between reconstructed and original point cloud increases
Solution Approach 1:
The patent segments the point cloud data into multiple patches, processes each patch independently through the neural network, and then fuses them together. This segmentation approach allows the system to handle large-scale point cloud data efficiently while maintaining high reconstruction accuracy in each local region, effectively resolving the contradiction between compression and precision.
Solution Approach 2:
The patent introduces a neural network as an intermediary between the compressed representation and the final reconstructed point cloud. This neural network learns the residual differences and uses them to refine the reconstruction, acting as a mediator that recovers lost information without requiring full transmission of original data, thus improving precision while maintaining compression benefits.
2Device complexity
If basic reconstruction is performed on original point cloud, then processing complexity is reduced, but quality and visual effect deteriorate
Solution Approach 1:
The patent performs preliminary processing by extracting patches from the point cloud before neural network processing. This pre-organization of data into manageable patches simplifies the subsequent neural network operations and facilitates efficient parallel processing, reducing overall complexity while enabling quality enhancement through learned residual refinement.
3Manufacturing precision
If neural network processing is applied to enhance attribute quality, then reconstruction accuracy is improved, but computational resource consumption increases
Solution Approach 1:
By dividing the point cloud into patches, the system can process smaller subsets of data through the computationally intensive neural network, enabling efficient resource utilization. The segmented approach allows for parallel processing and reduces the memory footprint, making high-accuracy reconstruction feasible with limited computational resources.
Solution Approach 2:
The patent applies neural network processing selectively to residual information rather than the entire point cloud data. This partial action approach focuses computational resources on the critical task of recovering lost information, achieving significant quality improvement while minimizing unnecessary computational overhead.
Data Source
AI summary
A coding method, an encoder, a decoder, and a readable storage medium are provided. The method includes the following. A reconstructed point set is determined based on a reconstructed point cloud, where the reconstructed point set includes at least one point. Geometry information and a reconstructed value of an attribute to-be-processed of a point in the reconstructed point set are input into a preset network model, and a processed value of the attribute to-be-processed of the point in the reconstructed point set is determined based on the preset network model. A processed point cloud corresponding to the reconstructed point cloud is determined according to the processed value of the attribute to-be-processed of the point in the reconstructed point set.