No-reference geometry quality assessment for colorless point clouds via list-wise rank learning

TL;DR

This work tackles the challenge of no-reference geometry-only quality assessment for colorless point clouds. It introduces LRL-GQA, a framework combining a multi-scale Patch-based Geometry Quality Assessment Network (GQANet) with a List-wise Rank Learning Network (LRLNet) trained on a large ranking dataset (LRL) of 52,200 geometrically distorted samples, plus a pseudo-MOS variant (LRL-PMOS) for scoring. The MPFE module is pre-trained via distortion-level classification, and the entire system is trained with a list-wise loss ($listMLE$) to directly optimize ranking while enabling absolute scores after fine-tuning. Experimental results show that LRL-GQA achieves ranking performance close to or surpassing full-reference metrics and yields competitive absolute quality scores after fine-tuning, highlighting its practical potential for geometry-only QoE assessment in 3D applications. The approach offers a scalable path to no-reference GQA, addressing data scarcity and enabling robust evaluation for tasks like reconstruction and compression in colorless point clouds.

Abstract

Geometry quality assessment (GQA) of colorless point clouds is crucial for evaluating the performance of emerging point cloud-based solutions (e.g., watermarking, compression, and 3-Dimensional (3D) reconstruction). Unfortunately, existing objective GQA approaches are traditional full-reference metrics, whereas state-of-the-art learning-based point cloud quality assessment (PCQA) methods target both color and geometry distortions, neither of which are qualified for the no-reference GQA task. In addition, the lack of large-scale GQA datasets with subjective scores, which are always imprecise, biased, and inconsistent, also hinders the development of learning-based GQA metrics. Driven by these limitations, this paper proposes a no-reference geometry-only quality assessment approach based on list-wise rank learning, termed LRL-GQA, which comprises of a geometry quality assessment network (GQANet) and a list-wise rank learning network (LRLNet). The proposed LRL-GQA formulates the no-reference GQA as a list-wise rank problem, with the objective of directly optimizing the entire quality ordering. Specifically, a large dataset containing a variety of geometry-only distortions is constructed first, named LRL dataset, in which each sample is label-free but coupled with quality ranking information. Then, the GQANet is designed to capture intrinsic multi-scale patch-wise geometric features in order to predict a quality index for each point cloud. After that, the LRLNet leverages the LRL dataset and a likelihood loss to train the GQANet and ranks the input list of degraded point clouds according to their distortion levels. In addition, the pre-trained GQANet can be fine-tuned further to obtain absolute quality scores. Experimental results demonstrate the superior performance of the proposed no-reference LRL-GQA method compared with existing full-reference GQA metrics.

Figures (10)

• Figure 1: Snapshots of some reference point clouds in the LRL dataset.
• Figure 2: Snapshots of point clouds degraded by different types of distortions.
• Figure 3: Overview of the proposed LRL-GQA framework. The LRL dataset contains ranked distorted point clouds for training and testing. The LRL-GQA consists of a geometry quality assessment network (GQANet) and a list-wise rank learning network (LRLNet). Based on the LRL dataset, the LRL-GQA can be trained to rank the list of degraded point clouds, as well as calculate absolute quality scores after fine-tuning.
• Figure 4: Overview of the GQANet framework. Given an input point cloud $\mathcal{P}$, the GQANet outputs its model quality index $I$.
• Figure 5: Structure of the multi-scale patch feature extraction (MPFE) module in the GQANet. For each patch $P_i$, the MPFE module employs edge convolution to extract its multi-scale quality-aware features $F_i$.