3D Gaussian Splatting Archive

FUNES 3D Gaussian Splatting Cultural Heritage Collection

A curated collection of four flagship archaeological and cultural heritage sites reconstructed with 3D Gaussian Splatting and packaged in PlayCanvas compressed .sog format for interactive browser inspection and spatial preservation research.

Flagship Radiance Fields

Included 3D Gaussian Splatting Scenes

Each flagship site includes an interactive full-screen browser viewer, its canonical archive record, and the compressed .sog distribution file.

Open interactive gallery →

.sog Packaging & Tooling Interoperability

Each scene is distributed in PlayCanvas Spatially Ordered Gaussians (.sog) format, compressing Morton-ordered Gaussian attributes into compact WebP textures plus a meta.json manifest for fast WebGL2 and WebGPU streaming.

# Convert .sog to standard .ply
npm install -g @playcanvas/splat-transform
splat-transform scene.sog scene.ply

Need raw multi-view photographs and COLMAP camera trajectories for training instead of viewing assets? Use the FUNES Open 3D Architecture Dataset.

Cite This Dataset

Switch between BibTeX, APA, and RIS formats to cite this resource in publications or reference managers.

@misc{FUNES_3dgs_cultural_heritage_2026,
  author       = {FUNES.WORLD Cultural Heritage Team},
  title        = {{FUNES 3D Gaussian Splatting Cultural Heritage Collection}},
  year         = {2026},
  publisher    = {FUNES.WORLD},
  version      = {1.0},
  url          = {https://funes.world/datasets/3dgs-cultural-heritage},
  note         = {License: CC BY-NC 4.0; Repository: https://funes.world/3dgs}
}

Technical Reference

Frequently Asked Questions

Practical guidance on coordinate conventions, evaluation protocols, file conversion, and licensing boundaries.

Why are these four heritage scenes packaged as .sog files instead of raw .ply checkpoints, and what trade-offs apply?

Developed by PlayCanvas, the Spatially Ordered Gaussians (.sog) format packs Morton-ordered Gaussian positions, scales, quaternion rotations, opacities, and clustered spherical harmonics palettes into lossless WebP attribute textures with a meta.json manifest inside a single compact archive—typically 15 to 20 times smaller than uncompressed .ply files for fast browser delivery. Because coordinate quantization and codebook clustering are lossy relative to raw float32 training checkpoints, .sog assets are engineered for real-time spatial inspection rather than resuming exact floating-point optimization.

How can I inspect or convert a .sog scene to standard .ply tooling?

You can inspect or convert any of the four .sog files (Pompeii, The Hanging Sculptures of the Xiaoxitian, Ani Ruins, and Van Castle) using the open-source PlayCanvas CLI @playcanvas/splat-transform (for example, splat-transform scene.sog scene.ply) or by opening the file in SuperSplat. Converting .sog to .ply exposes standard per-Gaussian positions, scales, rotations, opacities, and spherical harmonics for downstream analysis, though quantization applied during .sog packaging cannot restore pre-compression float32 precision.

Do these four .sog scenes include raw training photographs or COLMAP camera trajectories for benchmarking?

No. This collection publishes reconstructed viewing assets with guided camera bounds paired with canonical FUNES heritage records (/entities/...), rather than raw multi-view training frames, COLMAP sparse poses, or train/test splits. Researchers looking for multi-view photography and COLMAP Structure-from-Motion camera geometry to train or benchmark 3D Gaussian Splatting algorithms should use the FUNES Open 3D Architecture Dataset (/datasets/open-3d-architecture).

Does interactive playback require WebGPU, or do the scenes run in WebGL2 browsers?

All four scenes render directly through the PlayCanvas and SuperSplat browser runtime across both WebGL2 and WebGPU, allowing visitors on standard desktop and mobile hardware to explore the reconstructions without plugins. Browsers with WebGPU enabled accelerate compute-heavy sorting and editing workflows in newer SuperSplat pipelines, while WebGL2 viewers unpack the .sog attribute textures via standard texture sampling.

What scholarly and derivative uses are permitted for these .sog models under CC BY-NC 4.0?

Under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0), researchers, educators, and heritage conservationists may embed references, render figures for academic papers, or convert scenes for non-commercial spatial analysis with attribution to FUNES and the corresponding archive record. Commercial exhibitions, paid virtual experiences, and commercial generative or 3D model training using these assets require separate permission from FUNES.