Ahmed Doghri Logo Image
Ahmed Doghri

splatgrade

Thirty-six Gaussians earn a D: two giants, three duplicate pairs, four low-opacity points, and two floaters are localized before rendering.

splatgrade Gaussian point cloud inspection ring
splatgrade working browser demo

Rendered Quality Hides Scene Defects

A camera path can conceal giant primitives, duplicates, floaters, weak opacity, or unstable bounds until the scene reaches another renderer or viewpoint.

splatgrade parses real ASCII PLY fields, resolves common log-scale and opacity encodings, and reports robust scene statistics plus localized findings.

The Demo

The 36-Gaussian fixture receives grade D and score 49. It contains two giant primitives, three duplicate pairs, four low-opacity Gaussians, and two floaters.

The gate catches structural artifact risks. Perceptual quality still depends on view coverage, appearance, renderer settings, and human inspection.

Research Basis

Current 3DGS quality research emphasizes view-dependent and structural distortions that conventional image metrics can miss.

Read the 3D Gaussian Splatting quality assessment.

Tools Used

Python
3D Gaussian Splatting
PLY
Robust Statistics
Web Canvas
Docker
GitHub Actions CI