nianticlabs/mickey
A CVPR 2024 paper introducing MicKey, a differentiable pipeline that regresses keypoint positions in camera space and recovers metric relative poses from 2D image correspondences.

MicKey is a feature detection pipeline that regresses keypoint positions in camera space using descriptor matching to establish metric correspondences. From these correspondences, the system recovers metric relative camera poses in an end-to-end differentiable manner. The model is trained using differentiable pose optimization with only image pairs and ground truth relative poses as supervision. It uses RANSAC-based robust estimation alongside the learned components.
Frequently asked
- What is nianticlabs/mickey?
- A CVPR 2024 paper introducing MicKey, a differentiable pipeline that regresses keypoint positions in camera space and recovers metric relative poses from 2D image correspondences.
- Is mickey open source?
- Yes — nianticlabs/mickey is an open-source project tracked on heatdrop.
- What language is mickey written in?
- nianticlabs/mickey is primarily written in Python.
- How popular is mickey?
- nianticlabs/mickey has 592 stars on GitHub.
- Where can I find mickey?
- nianticlabs/mickey is on GitHub at https://github.com/nianticlabs/mickey.