.png){kind=icon}
.png){kind=icon}
Return
COPP-Net: No-Reference Point Cloud Quality Assessment via Weighted Patch Quality Prediction
.png){kind=icon}
PRE
AI
DOI：10.1109/TBC.2025.3597096.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As 3D vision applications relying on point clouds rapidly develop, point cloud quality assessment (PCQA) has emerged as a significant research area. When observing a point cloud, people typically rotate it to different viewpoints to examine local details from various angles, ultimately synthesizing the overall quality score of the point cloud. In this process, different parts of the point cloud have varying impacts on the overall quality. However, existing PCQA methods often overlook the influence of local quality variations across different regions of the point cloud. To address the imbalance in quality distribution, we introduce COPP-Net, a no-reference point cloud quality assessment (NR-PCQA) method equipped with the capability for local area correlation analysis. Specifically, we segment the point cloud into multiple patches and enhance PointNet++ to generate accurate texture and structure features for each patch. These features are then combined to predict the quality of each patch. Subsequently, we conduct aggregation analysis on the features of all patches using the correlation analysis (CORA) network based on Transformer to determine correlation weights. Finally, we calculate the overall quality score by combining the predicted quality and correlation weights of all patches. Through comparisons with the latest state-of-the-art NR-PCQA models, as well as a series of tests on different distortion types, cross-dataset validation, and time complexity analysis, the high performance of COPP-Net is verified. The available source code for the proposed COPP-Net can be found at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/philox12358/COPP-Net</uri>.
Keywords:
Point cloud quality assessment
point convolution
deep learning
Journal
IF:
4.8
Papers:
2.1K
Citations:
3.0K
View more organizations