gso-objects / README.md
suvadityamuk's picture
suvadityamuk HF Staff
catalog: add a Mesh column for the 3D dataset viewer
1c8a9a6 verified
|
Raw History Blame Contribute Delete
2.69 kB
metadata
license: cc-by-4.0
pretty_name: Google Scanned Objects, one object per folder
tags:
  - 3d
  - mesh
  - glb
  - obj
  - google-scanned-objects
  - trellis
size_categories:
  - 1K<n<10K
configs:
  - config_name: catalog
    data_files: catalog.parquet

Google Scanned Objects, one object per folder

The 1030 Google-authored scans of Google Scanned Objects, re-hosted from hugging-3d/google-scanned-objects so that every object can be downloaded on its own. Browse them, build subsets and export them in the GSO subset builder Space.

Layout

path what
objects/<id>/model.obj, model.mtl, texture.png original mesh, material and 4K texture, byte-for-byte
objects/<id>/model.glb normalized self-contained GLB (AABB centered, longest edge 1.0)
objects/<id>/metadata.json source metadata: name, description, category, brand, license, GLB stats, original size in meters
thumbnails/<id>/0.jpg ... 4.jpg the five original renders
previews/<id>.webp, previews/<id>.glb small grid tile and a 1024 px-texture GLB for web viewers
catalog.parquet one row per object (see below)

Download some objects

# one object, every file
hf download hugging-3d/gso-objects --repo-type dataset --include "objects/Threshold_Bead_Cereal_Bowl_White/*" --local-dir gso
# only the GLBs of every shoe
python -c "import pandas as pd; df = pd.read_parquet('hf://datasets/hugging-3d/gso-objects/catalog.parquet'); \
print(' '.join(f'objects/{i}/model.glb' for i in df[df.category == 'Shoe'].object_id))" \
  | xargs hf download hugging-3d/gso-objects --repo-type dataset --local-dir gso

Catalog columns

object_id, mesh (the object's lightweight GLB, a datasets.Mesh column that the dataset viewer renders in 3D), name, description, category, brand, group (colorway group, used for leak-free splits), vertices, faces, is_watertight, size_x_cm/size_y_cm/size_z_cm (real-world extents), max_dim_cm, flatness (shortest / longest extent), texture_px, obj_sha256 (the key of the object's TRELLIS.2 latents), and byte sizes of each file.

License and attribution

"Google Scanned Objects", Copyright 2020 Google LLC, licensed under CC BY 4.0. Credit Google Scanned Objects when you use these files. Only the Google-authored scans are included; files are unmodified except the derived GLBs and previews.