CAE-LO: LiDAR Odometry Leveraging Fully Unsupervised Convolutional Auto-Encoder for Interest Point Detection and Feature Description

@article{yin2020caelo,
    title={CAE-LO: LiDAR Odometry Leveraging Fully Unsupervised Convolutional Auto-Encoder for Interest Point Detection and Feature Description},
    author={Deyu Yin and Qian Zhang and Jingbin Liu and Xinlian Liang and Yunsheng Wang and Jyri Maanpää and Hao Ma and Juha Hyyppä and Ruizhi Chen},
    journal={arXiv preprint arXiv:2001.01354},
    year={2020}
}
@article{
    title={Interest Point Detection from Multi-Beam LiDAR Point Cloud Using Unsupervised CNN},
    author={Deyu Yin, Qian Zhang, Jingbin Liu, Xinlian Liang, Yunsheng Wang, Shoubin Chen, Jyri Maanpää, Juha Hyyppä, Ruizhi Chen},
    journal={IET Image Processing},
    year={2020}
}

See the rankings in KITTI, our method's name is "CAE-LO".

Thank you all for the interest. Now it's time to release the source code. But, currently I have some emergency personal problems to deal and I even told some people "I will release the code soon", so I decided to opensouce the code without fully optimization. Although the quality of the code is low and you may encount a problem when you run BatchPreprocess.py, the whole structure and the codes are readable and workable. And I think all the errors you may get will be easy to solve.

For the newers, I suggest you to wait for my formal release.

1. Basic enviornments for python3 and keras. No worries. I was a newer at that time, which is also part of the reason of why the code's quality is low.
2. Dirs.py to modify dictionaries.
3. BatchProcess.py to do batch processings on projecting PC to spherical rings and getting keypts by using multi-thread.
4. Voxel.py to project PC into multi-solution voxel model and basic functions about multi-resolution model.
5. SphericalRing.py to do basic function about spherical ring model, importantly for getting keypts by using trained model and cupy.
6. You can try Match.py to see some demos using trained models
7. PoseEstimation.py to generate initial odometry.
8. RefinePoses.py to generate refined odometry based on extended interest points and ground normals. (The code for generating ground normals is currently commented. Uncomment it if you want to use.)
9. There are some other test codes I tried to find other solutions. You may get some new ideas I hope. I will remove them in the formal release.

1. Generated interest points and features for sequence 00 and 01 can be found in GoogleDrive.
2. The extracting voxel patches part is slow, I will try to do some optimization.
3. The data arragement format is simple. Just serveral folders of "KeyPts", "Features", "InliersIdx", "SphericalRing", etc.
4. If you have any problems or confunsions, please post them in ISSUES or contact me by email. I am not sure if I can answer them quickly, but I will try my best.