alecjacobson/threedscans

Three D Scans

A mirror of threedscans.com, the archive of high-resolution 3D scans of museum objects initiated in 2012 by artist Oliver Laric.

134 meshes (10.6 GB) covering antiquities, classical and 19th-century sculpture, anatomical casts, and natural-history specimens, scanned in collaboration with museums across Europe.

This mirror exists so the collection can be fetched programmatically and cited stably. All credit for the scans belongs to Oliver Laric and the holding institutions.

Contents

Files	134 meshes, 10.6 GB
Formats	78 .stl (5.5 GB) and 56 .obj (4.8 GB), uncompressed
Metadata	metadata.json — per-object title, period, material, holding institution, scan year, scan method, source URL
Loader metadata	metadata.jsonl — the same fields keyed by filename, which datasets joins onto each .stl row

Mesh contents are mirrored exactly as served by threedscans.com — no re-meshing, re-encoding, or format conversion. Meshes are raw scan output: many are non-manifold or have boundary edges, which is often the point if you are testing geometry-processing code.

The site serves most objects as .zip. Those archives are unpacked here and the meshes stored bare, so the file you download is the file you use. The mesh bytes inside are untouched; what is dropped is the zip container and the __MACOSX/._* resource-fork entries macOS left in 32 of the archives.

This matters for disk: the archives expand from 5.3 GB to 10.6 GB, so consuming the zipped form meant storing both. Bare files are read straight from the Hub snapshot cache, so the data occupies disk once.

Filenames

A file is named for its object, with the extension lower-cased: Siren-1.stl, Actaeon.obj. Three cases are worth knowing about:

• Two distinct objects are both served as Hermanubis.stl.zip — one from the Vatican Museums, one from the Museo Archeologico dei Campi Flegrei a Baia. The Vatican object is prefixed with its upload date: 2017-01_Hermanubis.stl.
• Where the source collides in the same way, a -N suffix is appended to the stem: Statue-Provost_1M_Poly-1.stl, Xenophora-1.obj.
• Names inside the archives were unreliable — 25 contained spaces and several were stored in non-UTF-8 encodings — so names come from the archive, not its member.

Use the mesh_files field in metadata.json rather than deriving names from URLs; it always records the name as stored here. The files field is retained alongside it and records the original archive name the site serves, paired positionally with downloads.

Usage

from huggingface_hub import snapshot_download

path = snapshot_download(repo_id="alecjacobson/threedscans", repo_type="dataset")

Or a single object, loaded directly — no unzip step:

import trimesh
from huggingface_hub import hf_hub_download

f = hf_hub_download("alecjacobson/threedscans", "Siren-1.stl", repo_type="dataset")
mesh = trimesh.load(f)

datasets also works, and carries the attribution with each mesh:

from datasets import load_dataset          # requires datasets >= 5

ds = load_dataset("alecjacobson/threedscans", split="train")
row = ds[0]
row["mesh"]                                # trimesh object
row["title"], row["location"]              # 'Hermanubis', 'Vatican Museums'

Rows are backed by paths into the Hub snapshot cache, so nothing is copied or extracted a second time. metadata.jsonl supplies the title, artist, period, material, location, scanned, scanner, and source_url columns, joined on filename — so a decoded row still knows which object it is and who holds it.

Two limits:

• This path covers the 78 .stl files. datasets routes meshes through its meshfolder builder, which handles .glb, .ply, and .stl; .obj is not mapped to any builder. Reach the 56 .obj files through hf_hub_download as above, using metadata.json for their attribution.
• datasets gained mesh support in v5. On older versions neither the Mesh feature nor meshfolder exists, and load_dataset will not recognise these files; use snapshot_download, which has no such requirement and covers all 134.

metadata.json maps each source page to the files it provides:

{
  "url": "https://threedscans.com/uncategorized/siren/",
  "title": "Siren",
  "downloads": ["https://threedscans.com/wp-content/uploads/2025/10/Siren-1.stl.zip"],
  "files": ["Siren-1.stl.zip"],
  "mesh_files": ["Siren-1.stl"],
  "meta": {
    "artist": "Unknown",
    "period": "4th century CE",
    "material": "Marble",
    "location": "National Archaeological Museum Athens",
    "scanned": "2025",
    "scanner": "Photogrammetry and digital sculpting"
  }
}

Licensing and provenance

Read this before reusing the scans.

threedscans.com itself publishes no license text. As of July 2026 there is no licensing or terms statement on the homepage, the Info page, or any item page.

The permission relied on here is Laric's own statement on his companion project Lincoln 3D Scans, run with The Collection in Lincoln, which states that all models can be downloaded and used without copyright restrictions. Press coverage of threedscans.com reports the same intent for the wider archive: Laric asserts no copyright over the scans and intends them to circulate freely.

Two limits worth being precise about:

1. That explicit statement is published for the Lincoln project. It is strong evidence of intent for the whole archive, but threedscans.com carries no equivalent text of its own.
2. It is a statement of no restrictions, not a formal license instrument. This mirror therefore does not tag the data CC0 or PDDL — those are dedications only the rights holder can make, and Laric has not made one in those terms. Nothing here grants you rights beyond what Laric has stated.

The underlying objects are overwhelmingly public-domain works. Note that some jurisdictions may recognise rights in a 3D scan of a public-domain work, and holding institutions may impose their own conditions on commercial reuse.

If you reuse a scan, credit the project and the holding institution:

Scan by Oliver Laric, threedscans.com. Object: Siren, 4th century CE, marble, National Archaeological Museum Athens.

metadata.json carries the institution for every object so this can be done per-file.

Holding institutions

Albertina, Kunsthistorisches Museum and Theater Museum (Vienna); Musée Guimet, Musée des Monuments français, the Louvre, Dépôt des sculptures de la Ville de Paris, Musée Carnavalet (Paris); Museum Romanité (Nîmes); Parco Archeologico di Pompei; The Collection and Usher Gallery (Lincoln); Museo Archeologico Nazionale di Firenze; KODE Artmuseums (Bergen); National Archaeological Museum Athens; among others.

The project has been supported by Lafayette Anticipation, Secession Vienna, The Contemporary Art Society London, and Entrée Bergen.

Takedown

This is an unofficial mirror. If you are Oliver Laric or a holding institution and want material removed or the terms restated, open a discussion on this dataset and it will be taken down promptly.

Source code

Scraper and metadata harvester: github.com/alecjacobson/threedscans-scraper