Back to Blog
SEO

We Just Released Our Best Dexterous Hand Dataset – 41,160 High-Quality Frames for Humanoid Robotics

April 13, 20263 min read

We are excited to announce the release of our strongest Dexterous Hand Movements dataset to date.

This new batch delivers 41,160 high-quality frames extracted from 50 carefully selected videos, specifically optimized for training dexterous manipulation in humanoid robots.

Dataset Highlights

50 professional videos

41,160 HQ exported frames (including augmentations)

10,290 accepted frames before augmentation

Acceptance rate: 52.62%

Mean Quality Score: 0.9823 — one of our highest to date

Mean Landmark Visibility: 0.9646

Advanced Features Included

21-point MediaPipe Hands keypoints

Wrist-relative coordinates (wrist_position + wrist_relative_keypoints) — critical for accurate robot retargeting

Rich dexterous_hand block: finger angles, grip type proxies (pinch, precision, power, open, neutral), tip velocities

Motion Intelligence v3 (hand domain): hand activity classification, grip change detection, wrist & finger speeds

Gaussian temporal smoothing (window=5 on x/y/z)

Smart augmentations with wrist as pivot point

Quality & Transparency

Every frame has been strictly filtered and evaluated. You’ll receive full quality reports, per-video statistics, rejection logs, and a clean, well-documented structure (JSONL + per-video splits + SHA256 checksums).

This dataset is ideal for:

Dexterous manipulation and in-hand object handling

Imitation learning and teleoperation

Humanoid robot hand control (G1, GR-1, Walker, H2, etc.)

Gesture recognition and AR/VR applications

Fine motor skill training

Price: $440 (one-time commercial license)

Ready to try it?

Download the free sample or purchase the full dataset directly from our store. https://github.com/Alaaharoun/QualityVision-Motion-Dataset-Samples/tree/main/bulk_9_videos_933d5ad5_dataset

We also offer weekly supply packages up to 140,000+ frames per month and custom extraction services tailored to your exact requirements.