image imagewidth (px) 1.92k 3.84k |
|---|
tags: - 3d - human - photogrammetry - avatar - pose-estimation - digital-human - body-scan - dataset pretty_name: HumanAlloy 3D Human Scan Dataset size_categories: - 1K<n<10K
HumanAlloy 3D Human Scan Dataset
Dataset Description
HumanAlloy is a large-scale library of real human subjects captured with a professional 180-camera photogrammetry rig in controlled studio conditions. Each subject produces a clean, consistent full-body 3D mesh with realistic surface detail.
The dataset is the result of over 6,000 scanning sessions conducted between 2015 and 2022 at a dedicated photogrammetry studio in the Netherlands. It was developed as part of a professional scanning operation, later acquired by Snap Inc., and relaunched independently in 2023.
- Homepage: https://humanalloy.com
- Licensing inquiries: human@humanalloy.com
- AI training license info: https://humanalloy.com/pages/ai-training-data-licensing
Dataset Summary
| Property | Value |
|---|---|
| Total subjects | 4,700+ |
| Capture method | 180-camera photogrammetry rig |
| Geometry | Cleaned, decimated full-body mesh |
| Texture | 2K diffuse map per subject |
| Formats | OBJ, FBX |
| Raw source photos | ~130 per subject (available separately) |
| Model releases | Signed for all subjects |
| License type | Non-exclusive data licensing agreement |
Dataset Structure
Each subject is delivered as:
- Full-body mesh in OBJ and/or FBX format
- Single 2K diffuse texture map
- Preview renders (front, side, three-quarter)
Raw scan data (high-polygon mesh + ~130 calibrated source photographs per subject) is available as an add-on tier for teams that require source material for custom processing pipelines.
Supported Tasks
This dataset is suitable for:
- 3D human reconstruction — full-body geometry from real subjects
- Pose estimation — neutral and natural standing poses across a large population
- Avatar generation — high-quality geometry as training reference for generative models
- Clothing and apparel modeling — wide variety of outfit types captured
- Digital human development — diversity across age, gender, ethnicity, and body type
- Neural radiance fields (NeRF) / 3DGS — source photos available for view-synthesis training
Data Collection
Subjects were recruited and scanned at a professional studio in the Netherlands between 2015 and 2022. All subjects signed a model release agreement under Dutch law covering:
- Portrait rights
- Biometric data recorded during the scan session
- Third-party licensing and sublicensing
The scanning system used a custom-built 180-DSLR rig with controlled lighting and simultaneous capture. Reconstruction was performed in RealityCapture. Post-processing includes mesh cleanup, decimation, UV unwrapping, and texture baking.
Dataset Diversity
The library covers a broad range of:
- Gender — male and female subjects
- Age — from young adults to elderly
- Ethnicity — diverse representation across Western European population with international subjects
- Body type — slim, average, and larger builds
- Clothing — casual, business, sportswear, formal, seasonal outerwear
Licensing
The standard archviz license available on the HumanAlloy store does not cover AI training, generative model development, or dataset use.
A separate non-exclusive data licensing agreement is required. Licenses are available for:
- Internal AI and ML model training
- Research and development
- Commercial product development
For volume pricing and licensing inquiries: human@humanalloy.com
More information: https://humanalloy.com/pages/ai-training-data-licensing Sample meshes available for download: https://we.tl/t-rthHnXHEfza1XuX5
Citation
If you use this dataset in your research, please cite:
@dataset{humanalloy2023,
author = {Bethe, Tristan},
title = {HumanAlloy 3D Human Scan Dataset},
year = {2023},
publisher = {Bethe Beheer B.V.},
url = {https://humanalloy.com}
}
About
HumanAlloy was founded by Tristan Bethe, a 3D scanning specialist with 25+ years in the industry. The scanning operation was acquired by Snap Inc. in 2022 and relaunched independently in 2023. The dataset has been licensed to research institutions and technology companies for AI development projects.
- Downloads last month
- 8