Sample Data

This page covers the repo's own bundled/downloadable sample-data/ source files (curl commands, local conversion) plus open-data 3D Tiles for exercising tileset-level tools. For what's wired directly into the live viewer's clickable preset pills, see Live demo datasets.

RAD (Gaussian Splat LoD)

# Coit Tower, San Francisco — SparkJS BhattLod, SH1, 50M splats (1.2 GB monolithic)
curl -L https://wlt-ai-cdn.art/spark-2.0/rad/coit-40m-sh1-lod.rad -o coit.rad

# Jin-Ai-Machi — streaming manifest only (88 KB header, ~1.1 GB data fetched on demand)
curl -L https://cdn.lilea.net/spark/jinaimachi/jinaimachi-lod.rad -o jinaimachi.rad

# Tastier RAD 500K splats
curl -L https://wlt-ai-cdn.art/tastier_rad_500/0524c1a1-abf2-4969-ae40-9981ee836536_500k-lod.rad -o tastier500k.rad

RAD format note: A .rad file can be either monolithic (all chunk data embedded) or a streaming manifest (88 KB JSON header only; chunk data fetched on demand via HTTP range requests). The converter detects streaming manifests and exits with a clear error — download the full monolithic file to convert. Run node scripts/dump-rad-header.js <file.rad> to inspect.

Streamed SOG (PlayCanvas LOD)

PlayCanvas example datasets (download the whole tree with sog-to-3dtiles/scripts/fetch-sog.mjs):

cd packages/sog-to-3dtiles
# Skatepark — 20 chunks, 4 LOD levels (~160 files)
node scripts/fetch-sog.mjs https://code.playcanvas.com/examples_data/example_skatepark_02/ ../sample-data/input/sog/skatepark
# Roman Parish — 131 chunks, 7 LOD levels (~790 files)
node scripts/fetch-sog.mjs https://code.playcanvas.com/examples_data/example_roman_parish_02/ ../sample-data/input/sog/roman-parish
# then convert:
node src/index.js ../sample-data/input/sog/skatepark ../sample-data/output/sog/skatepark-3dtiles-from-sog

Both are wired into the viewer preset bar (⬢ Skatepark / Roman Parish (SOG)) and the inspector (cd tools-inspector && npm run inspect:skatepark / inspect:roman-parish).

COPC (LiDAR Point Cloud)

# Autzen Stadium, Oregon — classified, 10.6M pts (77 MB). Well-known test file.
curl -L "https://s3.amazonaws.com/hobu-lidar/autzen-classified.copc.laz" -o autzen.copc.laz

# USGS 3DEP Texas Coast 2018 — 50 cm density (154 MB)
curl -L "https://data.opengeos.org/USGS_LPC_TX_CoastalRegion_2018_A18_stratmap18-50cm-2995201a1.copc.laz" -o texas.copc.laz
# Online viewer: https://lidar-viewer.gishub.org/?url=https%3A%2F%2Fdata.opengeos.org%2FUSGS_LPC_TX_CoastalRegion_2018_A18_stratmap18-50cm-2995201a1.copc.laz

# IGN LiDAR HD — Tour Eiffel / Paris area, tile LHD_FXX_0648_6863 (France, 100 MB)
curl -L "https://data.geopf.fr/telechargement/download/LiDARHD-NUALID/NUALHD_1-0__LAZ_LAMB93_KE_2025-06-06/LHD_FXX_0648_6863_PTS_LAMB93_IGN69.copc.laz" -o eiffel.copc.laz

# USGS 3DEP — New Jersey tile (requires SAS token or STAC browser)
# https://usgslidareuwest.blob.core.windows.net/usgs-3dep-copc/usgs-copc/USGS_Lidar_Point_Cloud_NJ_SdL5_2014_LAS_2015/copc/18TWL805090.copc.laz

Online COPC viewers (inspect before converting):

• viewer.copc.io — drag-drop or paste URL
• lidar-viewer.gishub.org — drag-drop or paste URL

More COPC sources:

• IGN France — https://cartes.gouv.fr/telechargement/IGNF_NUAGES-DE-POINTS-LIDAR-HD Full national LiDAR HD coverage at 10 pts/m². Search tile ID (e.g. LHD_FXX_0648_6863).
• USGS 3DEP STAC — https://planetarycomputer.microsoft.com/api/stac/v1/collections/3dep-lidar-copc Browser: https://radiantearth.github.io/stac-browser/#/external/planetarycomputer.microsoft.com/api/stac/v1/collections/3dep-lidar-copc

I3S (Esri SceneServer / SLPK)

I3S open-data samples (beyond Buildings_NewYork_v18.slpk, mesh-only) — SLPK + hosted-service pairs for both integrated mesh and point-cloud layers, from Esri's own i3s-spec sample data readme, the ArcGIS JS SDK 3D-object sample gallery, and the ArcGIS 3D Tiles samples experience (ArcGIS portal items — open each item.html link and use its "Download" button for the SLPK, or the REST service URL directly for streaming). All four verified live against /stream?format=i3s:

• Integrated mesh (works, fully textured) — SLPK: item 95a427c7a6ec4789b03c1a177366b54c (Rancho_Mesh_v18.slpk, 285 MB) · service: item 01eff699c8404a27a65e0877201136b4 → https://tiles.arcgis.com/tiles/z2tnIkrLQ2BRzr6P/arcgis/rest/services/Rancho_Mesh_v18/SceneServer. The SLPK link is ArcGIS Online's own download endpoint (.../content/items/<id>/data) — no .slpk anywhere in the URL (only in a Content-Disposition header), which the adapter didn't originally detect as an archive at all; both this and the texture support were fixed 2026-07-02, verified byte-identical (same JPEG at the same buffer offset) against both the SLPK and the SceneServer URL for the same underlying dataset.
• Point cloud (LEPCC-compressed — now decoded, both work) — SLPK: item 496552d059644b4892c51ad06bdba8e2 (Moro_Bay_LiDAR.slpk, 578 MB, 2,624 tiles) · service: item 908d6b986f314d51b1ff50b3bc321dfd → https://tiles.arcgis.com/tiles/z2tnIkrLQ2BRzr6P/arcgis/rest/services/Moro_Bay_LiDAR/SceneServer (13,775 tiles). Position decodes correctly (checksum + extent-verified); RGB doesn't render yet — it's a separate LEPCC blob type not yet ported, and this specific dataset's RGB attribute resource 404s server-side anyway despite being advertised in its schema. A second point-cloud sample (item fc3f4a4919394808830cd11df4631a54, BARNEGAT_BAY_LiDAR_UTM) is also LEPCC — every real ArcGIS point-cloud SceneServer sample found so far uses it, since it's Esri's own default point-cloud codec.
• Live ArcGIS SceneServer REST support added (2026-07-02) to test the "service" URLs above directly (previously the adapter only read a static 3dSceneLayer.json/SLPK tree, not the REST API's ?f=json-based resource paths) — see arcgisSceneServerReader() in i3s-live.js.
• Also referenced: portal item fc3f4a4919394808830cd11df4631a54 (point cloud) and 001bb7ee3ce44ae5a8a15bef72f4404a (integrated mesh) — not yet run against i3s-to-3dtiles//stream?format=i3s, listed here as further samples to pull in.

Open-Data 3D Tiles sample datasets

Cesium and partner organizations, plus a handful of public repositories, provide ready-to-use OGC 3D Tiles datasets for testing converters, viewers, and tools — as opposed to the bundled non-3D-Tiles source data above, these are already 3D Tiles, useful for exercising the implicit-to-explicit / upgrade tileset-level tools. Several are used by this project's Tools mode in the viewer.

CesiumGS open data (Ion) — https://cesium.com/platform/cesium-ion/content/

• Assets marked "open data" are publicly accessible with any Ion account token.
• Examples: Google Photorealistic 3D Tiles (assetId 2275207), Bing Imagery basemap.

More public 3D Tiles / 3DGS / OSM-3D endpoints — see the curated list in blosm#498 (3D Tiles + Gaussian-splat + city-model sources).

Esri I3S samples (for i3s-to-3dtiles) — loaders.gl I3S test data has both .slpk packages and raw REST/file-server layouts (3dSceneLayer.json + nodepages/ + nodes/ trees on disk — the converter reads those directly, no unzip). Mix of 3DObject/IntegratedMesh (mesh), PointCloud, and Draco-compressed layers (Draco/LEPCC are gated out). Local samples live in sample-data/input/I3S-SLPK/ (NYC buildings mesh ✓, Autzen point cloud → LEPCC, gated).

bertt/cesium_3dtiles_samples

https://github.com/bertt/cesium_3dtiles_samples

Hosted at https://bertt.github.io/cesium_3dtiles_samples/samples/.

Utrecht 3D

tileset.json: asset.version = "1.0"
root.implicitTiling:
  subdivisionScheme: QUADTREE
  subtreeLevels: 7
  subtrees.uri: subtrees/{level}_{x}_{y}.subtree   ← underscore separator
root.content.uri: content/{level}_{x}_{y}.b3dm

• Implicit 1.0 QUADTREE with b3dm tiles — requires both implicit-to-explicit (to unroll the quadtree into an explicit children tree) and b3dm→glb conversion per tile.
• The /3dtiles-tools?tool=implicit-to-explicit endpoint handles both in one pass: builds the explicit tree, then converts b3dm→glb on each tile request.
• No availableLevels in the tileset — the server walks subtrees until no child subtrees are found.

Delaware 1.1

tileset.json: asset.version = "1.1"
root.implicitTiling:
  subdivisionScheme: QUADTREE
  subtreeLevels: 3
  availableLevels: 4
  subtrees.uri: subtrees/{level}_{x}_{y}.subtree
root.content.uri: content/{level}_{x}_{y}.glb

• Fully spec-compliant 3D Tiles 1.1 QUADTREE implicit tileset with .glb tiles.
• implicit-to-explicit unrolls to an explicit tree; tiles are proxied as-is (already GLB).

CesiumGS/3d-tiles-samples

https://github.com/CesiumGS/3d-tiles-samples

Hosted at https://raw.githubusercontent.com/CesiumGS/3d-tiles-samples/main/1.1/ (raw GitHub, CORS-friendly, verified 200) — comprehensive reference samples for all 3D Tiles 1.1 features:

The SparseImplicit* samples are the canonical correctness tests for implicit-to-explicit: they exercise {"constant": 0} / {"constant": 1} sparse availability and cover both Morton-2D (QUADTREE) and Morton-3D (OCTREE) decode paths.

SparseImplicitOctree URL: https://raw.githubusercontent.com/CesiumGS/3d-tiles-samples/main/1.1/SparseImplicitOctree/tileset.json

Other open datasets used in this project

All three city mesh tilesets below use RGF93/Lambert-93 or WGS84 coordinates and are served with CORS — they work in viewer.html via the built-in city mesh presets.

Lille pyramid tileset

tileset.json: asset.version = "1.0"
Explicit tree (no implicitTiling)
Tile content: *.b3dm  (Batched 3D Model, 3D Tiles 1.0)

• Large photogrammetry mesh of Lille, France (10 cm resolution).
• Use the upgrade tool to serve b3dm tiles as GLB on-demand without pre-converting.
• The pyramid/ subdirectory is a sub-tileset of the full dataset.

For the datasets this repo's own converters read from (COPC/Potree/SOG/RAD/LCC/I3S sources, not already-3D-Tiles), see the sections above. For a tour of what's wired into the live viewer's preset pills specifically (including these same open-data 3D Tiles as its DIRECT presets), see Live demo datasets.