Reproduce “Perceptual Quality Assessment of 3D Gaussian Splatting: A Subjective Dataset and Prediction Metric” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhaolin Wan, Yining Diao, Jingqi Xu, Hao Wang, Zhiyang Li, Xiaopeng Fan, Wangmeng Zuo, Debin Zhao
This work introduces 3DGS-QA, the first subjective quality assessment dataset for 3D Gaussian Splatting (3DGS), and proposes GSOQA, a no-reference perceptual quality prediction model operating directly on native Gaussian primitives. GSOQA achieved a Pearson Linear Correlation Coefficient (PLCC) of 0.8213 and a Spearman Rank Order Correlation Coefficient (SRCC) of 0.7738, outperforming existing image, video, and 3D quality assessment methods.
Key insights
- Reconstruction distortions (e.g., reduced viewports) elicit a bimodal perceptual response, while synthesis distortions (e.g., noise) result in a more gradual, near-normal quality degradation.
- Directly operating on native 3D Gaussian primitives enables GSOQA to capture structural integrity and perceptual quality more effectively than methods relying on 2D projections or explicit geometric models.
- The Graph Attention Network (GAT) and attribute-aware attention pooling are essential for modeling relational dependencies among Gaussians and prioritizing perceptually salient regions, contributing to GSOQA's robust performance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.