HSRD-100: Free 3D Human Scan Dataset for AI, Research, and Digital Humans

At Digital Reality Lab, we specialize in high-quality 3D human scans for industries ranging from gaming and film to machine learning and product design. Today, we’re proud to release HSRD-100, a free dataset of 100 premium 3D human scans available on Hugging Face.

This dataset is part of our broader HumanScanRepository, one of the largest commercial collections of photogrammetry-based human scans in the world.

Whether you’re working on AI research, 3D avatars, pose estimation, or virtual fashion, HSRD-100 offers a high-fidelity resource to prototype, benchmark, and accelerate your projects.

What’s Included in HSRD-100

The HSRD-100 dataset is carefully designed to give researchers and developers a realistic sample of our much larger HumanScanRepository.

• 100 Scans

  • 10 individuals × 10 distinct poses.

  • Balanced across gender and diverse ethnic groups.

• High-Resolution 3D Meshes

  • Raw captures up to 5 million polygons.

  • Multiple Levels of Detail (LOD):

    • LOD0: ~1M polygons, 16k textures

    • LOD1: ~100k polygons, 8k textures

    • LOD2: ~10k polygons, 4k textures (ideal for real-time).

• Photorealistic Textures

  • 4K–16K resolution, captured with professional 112-camera photogrammetry rigs.

• Clothing & Pose Diversity

  • Neutral, walking, and expressive poses.

  • Outfits include casual wear, business attire, sportswear, and formal clothing.

• Rich Metadata

  • person_meta.json (age, sex, ethnicity, height, weight).

  • pose_meta.json (pose type, activity, clothing style).

  • Technical scan parameters for reproducibility.

• Source Data

  • RealityCapture projects.

  • Raw photographic sets.

  • Preview/cover images for quick browsing.

• Open License

  • CC BY-4.0 — free for both academic and commercial use with attribution.

👉 Download HSRD-100 on Hugging Face

Why Use HSRD-100?

1. Accelerate AI and ML Research

HSRD-100 provides a ground-truth dataset of 3D humans for:

• Pose estimation

• Mesh segmentation

• 3D reconstruction

• Neural rendering and generative AI

2. Develop Realistic 3D Avatars

Game studios, VR/AR developers, and VFX artists can use HSRD-100 as a base for avatar creation with clean topology and high-quality textures.

3. Test Synthetic Data Pipelines

Metadata enables structured augmentation, making it easy to filter scans by age, gender, ethnicity, or pose type.

4. Explore Digital Fashion & Virtual Try-On

The inclusion of diverse clothing styles makes the dataset ideal for early testing in virtual fitting rooms and apparel research.

5. Support Education & Prototyping

HSRD-100 is lightweight enough for universities, labs, and independent developers to integrate into teaching, prototyping, or exploratory research.

Limitations of a 100-Model Dataset

While HSRD-100 is valuable, it’s important to note its limitations:

• Only 10 individuals, which limits morphological and demographic diversity.

• Limited to 10 poses per subject; no extreme athletic, dance, or floor poses.

• Age range is 21–64; children and older adults are not included.

• Controlled studio conditions — scans are consistent but may not reflect noisy real-world data.

This makes HSRD-100 best for small-scale research and prototyping, not for training large-scale ML models.

Need More Data? Explore the Full HumanScanRepository (20,000+ Scans)

For researchers and organizations requiring scale and diversity, our full HumanScanRepository offers:

• 20,000+ high-quality scans with extensive demographic coverage.

• Thousands of unique poses across activities, gestures, and motion states.

• Rich clothing variety and larger representation of body shapes, ethnicities, and ages.

• Ready-to-use data for machine learning, computer vision, XR development, and product design.

📩 To access the full dataset, contact Digital Reality Lab at ivo@digitalrealitylab.com or through HumanScanRepository.com.

Ethical Use & Licensing

We emphasize responsible dataset use:

• Participants gave informed consent.

• Data is anonymized; no biometric identification intended.

• Not for surveillance, medical diagnosis, or sensitive biometric tasks.

• Attribution required under CC BY-4.0 license.