计算机科学
缩放
特征(语言学)
云计算
点(几何)
点云
人工智能
情报检索
数据挖掘
数学
哲学
语言学
几何学
石油工程
工程类
镜头(地质)
操作系统
作者
Jilong Wang,Wei Gao,Ge Li
出处
期刊:IEEE Transactions on Circuits and Systems for Video Technology
[Institute of Electrical and Electronics Engineers]
日期:2024-01-01
卷期号:: 1-1
标识
DOI:10.1109/tcsvt.2024.3362369
摘要
Point clouds offer a novel 3D data representation that has proven pivotal in immersive visual media applications involving human perception. Developing objective point cloud quality assessment (PCQA) methods is imperative, as they can substantially reduce human evaluation costs and drive advancements for visual perceptual experiences in point cloud related applications. Point cloud quality assessment without reference remains challenging. Previous PCQA methods predominantly employ a fixed perceptual distance and often overlook the variability in quality perceived from different viewpoints, which impedes their effectiveness in multiscale or multi-granularity feature extraction and learning, particularly for deep neural networks. The single fixed observation distance fails to capture the multi-resolution characteristics intrinsic to human perception. Addressing this gap, in this paper, we introduce a novel no-reference PCQA method (MOD-PCQA) that integrates multiscale features to enhance point cloud quality perception across diverse scales and granularities. MOD-PCQA pioneers a viewpoint aware feature learning framework, capable of capturing visual features across various granularity levels, from fine to coarse. Specifically, we process and project point clouds into images from different viewpoints. Then, we extract multi-scale features under corresponding perspectives through three branch networks. Finally, we design an alternate learning strategy to optimize the feature extraction network to continuously refine the learned feature information from both inter-scale and intra-scale perspectives. Comprehensive experiments conducted on the SJTU-PCQA and WPC databases validate the superiority of our proposed model over state-of-the-art PCQA methods. Our method achieves optimal performance on both benchmarks by a significant margin, which comprehensively validates its effectiveness for the challenging PCQA task. The source code will be available at https://openi.pcl.ac.cn/OpenPointCloud/MOD-PCQA [1].
科研通智能强力驱动
Strongly Powered by AbleSci AI