zc-alexfan/hold
A method that jointly reconstructs 3D articulated hands and objects from monocular video using neural networks without requiring pre-scanned object templates or 3D training data.
Not currently ranked — collecting fresh signals.
star history
HOLD is a CVPR 2024 research project that uses deep learning to simultaneously reconstruct articulated hands and objects from single-view RGB videos. The method leverages neural networks and PyTorch to achieve category-agnostic 3D reconstruction without assuming pre-scanned object templates or requiring 3D hand/object training data. It addresses hand-object interaction reconstruction tasks relevant to augmented and virtual reality applications.
Frequently asked
- What is zc-alexfan/hold?
- A method that jointly reconstructs 3D articulated hands and objects from monocular video using neural networks without requiring pre-scanned object templates or 3D training data.
- Is hold open source?
- Yes — zc-alexfan/hold is open source, released under the MIT license.
- What language is hold written in?
- zc-alexfan/hold is primarily written in Python.
- How popular is hold?
- zc-alexfan/hold has 485 stars on GitHub.
- Where can I find hold?
- zc-alexfan/hold is on GitHub at https://github.com/zc-alexfan/hold.