The Worse The Better: Content-Aware Viewpoint Generation Network for Projection-related Point Cloud Quality Assessment

TL;DR

This work addresses the sensitivity of projection-based point cloud quality assessment (PCQA) to viewpoint selection. It introduces a content-aware viewpoint generation network (CAVGN) that uses multi-scale geometric and texture features and a feature constraint mechanism to generate optimized viewpoints for each default projection region, guided by a self-supervised default-optimized viewpoint dataset (DOV) and a ranking network (SSVRN). Training leverages SSVRN to construct DOV without manual labeling, enabling CAVGN to produce content-reflective viewpoints. Across SJTU-PCQA, WPC, and LS-PCQA, CAVGN-generated viewpoints substantially improve both NR and FR/RR PCQA methods, with cross-dataset validation demonstrating robust generalization. The approach offers practical gains for projection-based PCQA pipelines, albeit with increased computational cost that the authors acknowledge and propose addressing in future work.

Abstract

Through experimental studies, however, we observed the instability of final predicted quality scores, which change significantly over different viewpoint settings. Inspired by the "wooden barrel theory", given the default content-independent viewpoints of existing projection-related PCQA approaches, this paper presents a novel content-aware viewpoint generation network (CAVGN) to learn better viewpoints by taking the distribution of geometric and attribute features of degraded point clouds into consideration. Firstly, the proposed CAVGN extracts multi-scale geometric and texture features of the entire input point cloud, respectively. Then, for each default content-independent viewpoint, the extracted geometric and texture features are refined to focus on its corresponding visible part of the input point cloud. Finally, the refined geometric and texture features are concatenated to generate an optimized viewpoint. To train the proposed CAVGN, we present a self-supervised viewpoint ranking network (SSVRN) to select the viewpoint with the worst quality projected image to construct a default-optimized viewpoint dataset, which consists of thousands of paired default viewpoints and corresponding optimized viewpoints. Experimental results show that the projection-related PCQA methods can achieve higher performance using the viewpoints generated by the proposed CAVGN.

Figures (6)

• Figure 1: Comparison of default viewpoints and optimized viewpoints generated by CAVGN.
• Figure 2: Overview of the proposed content-aware viewpoint generation network (CAVGN).
• Figure 3: Illustration of candidate viewpoints setting on the projection region $r_i$ under $N_v = 9$.
• Figure 4: Overview of the proposed self-supervised viewpoint ranking network (SSVRN).
• Figure 5: Comparison of default and generated viewpoints of hhi (a) and redandblack (b) in the SJTU-PCQA dataset.