This works intends to compare the result from the different pointfeature source & compare their performance.
Currently included:
- SuperPoint
- R2D2
- OpenCV-based
- ORB
- SIFT
- SuperGLUE
- OpenCV: BF-matcher
- AdaLAM from kernia
- Detector-Free
- LoFTR from kernia
- MAGSAC
- OnePose (3D to 2D matching)
The current architecture we used is based ZMQ socket. The reason why we should this is because this can help us to integrate different conda environment, python and c++. Make the whole structure easy to use.
We also provide python and C++ wrapper to integrate with your own projects.
- Add python client.
- Add CUDA-based data passing support.
- Add OnePose to this structure.
- Add OnePose method
- Add network client & wrapper