CalciferZh/minimal-hand
A deep learning solution for real-time hand motion capture from a single RGB camera, achieving over 100fps through neural networks for joint location and rotation estimation.
Not currently ranked — collecting fresh signals.
star history
This project provides a complete hand motion capture system using deep learning from a single webcam. It estimates 3D hand joint locations from monocular RGB images using DetNet, then predicts joint rotations from those locations using IKNet. The approach is robust to occlusion, hand-object interaction, fast motion, and varying scales and viewpoints. With pre-trained models, processing runs at approximately 8.9ms for detection and 0.9ms for inverse kinematics on a 1080Ti GPU.
Frequently asked
- What is CalciferZh/minimal-hand?
- A deep learning solution for real-time hand motion capture from a single RGB camera, achieving over 100fps through neural networks for joint location and rotation estimation.
- Is minimal-hand open source?
- Yes — CalciferZh/minimal-hand is open source, released under the MIT license.
- What language is minimal-hand written in?
- CalciferZh/minimal-hand is primarily written in Python.
- How popular is minimal-hand?
- CalciferZh/minimal-hand has 1k stars on GitHub.
- Where can I find minimal-hand?
- CalciferZh/minimal-hand is on GitHub at https://github.com/CalciferZh/minimal-hand.