fbpx
Skip to content Skip to footer
Hand-Object Tracking in 3D

HOT3D

A comprehensive egocentric dataset for 3D hand and object pose tracking, captured from Project Aria glasses. Features high-fidelity ground truth annotations and multi-modal sensor data.

WHAT IS IT?

A new benchmark dataset to better understand how humans use their hands

We use our hands to communicate with others, interact with objects, and handle tools. Yet reliable understanding of how people use their hands to manipulate objects remains a key challenge for computer vision research.

The HOT3D dataset and benchmark will unlock new opportunities within this research area, such as transferring manual skills from experts to less experienced users or robots, helping an AI assistant to understand user’s actions, or enabling new input capabilities for AR/VR users, such as turning any physical surface to a virtual keyboard or any pencil to a multi-functional magic wand.

Dataset Specifications

TOTAL SEQUENCE

92,408 hand-object interactions

Total Size

2.5TB

Video Resolution

1408 x 1408

Frame Rate

1408 x 1408

Recording Duration

800 minutes of egocentric recordings

Unique Objects

33 hand-held objects

Accurate ground-truth 3D poses of hands and objects

A set of small optical markers were attached to hands and objects and tracked using a professional motion-capture system. This ground truth enables training and evaluating methods for joint hand and object tracking.

High-fidelity 3D object models

To enable research on model-based object pose estimation, we provide high-fidelity 3D models of 33 diverse objects. Each model is captured with high-resolution geometry and PBR materials, using an in-house 3D scanner.
HOT-3D DATASET TOOLS

Comprehensive tools to load and visualize data easily

We provide python tools that enable researchers to interact with egocentric hands and objects tracking in 3D on multi-view image streams.

An API and code samples provide ways to easily access and visualize the image streams and high-quality ground-truth 3D poses and shapes of hands and objects.

Key Features

Synchronized multi-view egocentric videos from Project Aria glasses and Quest 3 VR headset
High-quality 3D pose annotations of hands and objects
Multi-view calibrated cameras
IMU and eye tracking data
Semantic segmentation masks
Object mesh reconstructions

Access HOT3D Dataset and accompanying Tools

If you are a researcher in AI or ML research, access the HOT3D Dataset and accompanying tools here.

    By submitting your email and accessing the HOT3D dataset, you agree to abide by the dataset license agreement and to receive emails in relation to the dataset.

    Downloads

    Full Dataset (Subscription Required)

    11 KB DOCX

    Sample Data

    11 KB DOCX

    Sample Annotations

    1,018 KB PDF

    Documentation

    1,018 KB PDF

    Citation

    @article{hot3d2024,
      title={HOT3D: Hand-Object Tracking in 3D},
      author={Dexterous Research Team},
      journal={arXiv preprint},
      year={2024}
    }