| Dataset | Focus | Formats | Annotations | Best For | License | Official |
|---|---|---|---|---|---|---|
| Objaverse‑XL | Diverse textured objects at scale | GLB, USDZ | Categories, license, metadata | Generative 3D, reconstruction | CC‑BY (check per object) | HF • Paper |
| Objaverse++ | Neural fields + physics notes | GLB, USDZ, NeRF, 3DGS | Neural/physics metadata | Neural rendering & simulation | CC‑BY (check per object) | HF • Paper |
| ShapeNet | CAD‑like taxonomy | OBJ/MTL (varies) | Consistent categories | Geometry benchmarks | Research terms | Site |
| Google Scanned Objects (GSO) | Clean real‑world scans | GLTF/GLB | Categories, scans | Realistic rendering | Dataset license | |
| ModelNet | Classic academic benchmark | OFF/PLY/OBJ (varies) | Categories | Baseline evaluation | Research terms | Princeton |
Notes: Always confirm licensing on the official pages. “Best for” and formats are generalized summaries.
Choosing the Right Dataset
- Need neural fields? Use Objaverse++.
- Need raw textured meshes at scale? Use Objaverse‑XL.
- Benchmarking geometry or taxonomy? ShapeNet/ModelNet.
- Realistic materials and scans? GSO.
Related Guides
NeRF vs 3D Gaussian Splatting • Hardware for Objaverse • Subsets & Metadata