DGTXxx/citygs-spark-assets

CityGS Spark Assets

This repository hosts the public model assets and build notes used by the CityGS SparkJS web viewer. It is intended as an asset repository for browser rendering and demo deployment, not as a raw training dataset.

The files are prepared for the CityGS/SparkJS website workflow: the website loads model files from public/models/, while the documentation under docs/ records the Mac Docker build/export path used for local packaging.

Repository Contents

models/
  mc_aerial_c36_full_23648k.compressed.ply
  mc_aerial_c36_light_50_vq_lod1_11824k.spz
  mc_aerial_c36_light_66_vq_lod2_8040k.spz
  mc_aerial_c36_light_75_vq_full_5912k.ply
  mc_aerial_c36_light_75_vq_full_5912k.spz
  mc_aerial_c36_light_75_vq_preview_*.ply
  mc_aerial_coarse_input.ply
  mc_aerial_coarse_iter30000_7479k.spz

docs/
  mac-spark-docker-build.md

The models/ directory contains MatrixCity Aerial model assets at several levels of detail. The .spz files are compact Spark-compatible assets for web loading, while the .ply files are larger Gaussian assets used for full, preview, or conversion workflows.

Quick Download

Install the Hugging Face CLI first:

python3 -m pip install -U huggingface_hub hf_transfer

Download all model assets into a website project:

HF_HUB_ENABLE_HF_TRANSFER=1 hf download DGTXxx/citygs-spark-assets \
  --repo-type dataset \
  --include "models/*" \
  --local-dir frontend/public

Expected local layout:

frontend/public/models/

Download only the Mac Docker build guide:

hf download DGTXxx/citygs-spark-assets \
  --repo-type dataset \
  --include "docs/mac-spark-docker-build.md" \
  --local-dir .

Use With the CityGS SparkJS Viewer

Place the downloaded models/ directory under the website's public asset directory:

frontend/public/models/

The viewer can then serve the assets from paths such as:

/models/mc_aerial_c36_light_75_vq_full_5912k.spz
/models/mc_aerial_c36_light_75_vq_preview_500k.ply

For Mac local packaging, download docs/mac-spark-docker-build.md and follow the Docker export workflow from the website project root. The current helper script path used by the project is:

./scripts/export-spark-docker.sh

File Notes

• *_preview_*.ply: reduced-size preview assets for faster loading and testing.
• *_full_*.ply and *.compressed.ply: larger assets for higher-quality views or conversion workflows.
• *.spz: compact Spark-compatible assets intended for web delivery.
• mc_aerial_coarse_*: coarse MatrixCity Aerial assets used by the demo and related conversion/testing workflows.

Scope and Limitations

This repository does not contain the original MatrixCity dataset, training images, training code, or complete reconstruction pipeline. It only contains derived model assets and deployment notes used by the web viewer.

The Hugging Face Dataset Viewer is disabled because these files are binary 3D assets rather than row-based tabular data.

License and Attribution

The repository is marked with license: other because the assets are derived from third-party research datasets and 3D Gaussian Splatting tooling. Use these files for research, demonstration, and project reproduction only, and respect the licenses and terms of the original datasets and upstream tools, including MatrixCity, CityGaussian, SparkJS, and related 3DGS projects.