Perceptual Quality Assessment of 3D Gaussian Splatting: A Subjective Dataset and Prediction Metric

TL;DR

This work addresses the gap in perceptual quality assessment for 3D Gaussian Splatting (3DGS) by introducing 3DGS-QA, the first subjective dataset for 3DGS with 225 distorted models across 15 object types, and a no-reference predictor GSOQA. The model directly analyzes native Gaussian primitives, employing geometric preprocessing, a Gaussian-MAE for semantic features, and a cascaded Graph Attention Network with attention pooling to predict perceptual quality, trained with a hybrid loss that enforces linearity and monotonicity with respect to MOS. Extensive benchmarking against IQA, VQA, and 3DQA baselines shows GSOQA consistently outperforms existing metrics, particularly under reconstruction and downsampling distortions, validating its robustness for 3DGS content. The dataset and code are publicly available to accelerate further research in perceptual quality assessment and rendering optimization for 3DGS content.

Abstract

With the rapid advancement of 3D visualization, 3D Gaussian Splatting (3DGS) has emerged as a leading technique for real-time, high-fidelity rendering. While prior research has emphasized algorithmic performance and visual fidelity, the perceptual quality of 3DGS-rendered content, especially under varying reconstruction conditions, remains largely underexplored. In practice, factors such as viewpoint sparsity, limited training iterations, point downsampling, noise, and color distortions can significantly degrade visual quality, yet their perceptual impact has not been systematically studied. To bridge this gap, we present 3DGS-QA, the first subjective quality assessment dataset for 3DGS. It comprises 225 degraded reconstructions across 15 object types, enabling a controlled investigation of common distortion factors. Based on this dataset, we introduce a no-reference quality prediction model that directly operates on native 3D Gaussian primitives, without requiring rendered images or ground-truth references. Our model extracts spatial and photometric cues from the Gaussian representation to estimate perceived quality in a structure-aware manner. We further benchmark existing quality assessment methods, spanning both traditional and learning-based approaches. Experimental results show that our method consistently achieves superior performance, highlighting its robustness and effectiveness for 3DGS content evaluation. The dataset and code are made publicly available at https://github.com/diaoyn/3DGSQA to facilitate future research in 3DGS quality assessment.

Figures (6)

• Figure 1: The source models of our database 3DGS-QA.
• Figure 2: Examples of distortion types in 3DGS models: (a) reduced viewpoints, (b) limited training, (c) point downsampling, (d) spatial noise, and (e) color perturbation.
• Figure 3: The subjective quality assessment interface.
• Figure 4: MOS distribution across reconstruction (red curve) and synthesis (blue curve) distortions in the dataset.
• Figure 5: MOS statistics across downsampling, Gaussian noise, and color noise distortions.