ltriess / kitti_scan_unfolding

:snake: Python Implementation for KITTI Scan Unfolding

KITTI Scan Unfolding

Python implementation of KITTI scan unfolding.

The code is currently not available on this repository. Feel free to contact me in case you have any questions.

We propose KITTI scan unfolding in our paper Scan-based Semantic Segmentation of LiDAR Point Clouds: An Experimental Study. If you use this code or the algorithm described in the paper, please make sure to cite this paper.

Fig. 1: Cylindrical point cloud projection: (a) Correcting for ego-motion leads to a projection that suffers from systematic point occlusions as some 3D points are projected into occupied pixels. Hidden points can not provide any information to the network and may not be accurately classified. (b) The scan unfolding method provides a dense projection without systematic discretization artifacts.

Citation

@inproceedings{triess2020iv,
    title = {{Scan-based Semantic Segmentation of LiDAR Point Clouds: An Experimental Study}},
    author = {Triess, Larissa T. and Peter, David and Rist, Christoph B. and Z\"ollner, J. Marius},
    booktitle = {IEEE Intelligent Vehicles Symposium (IV)},
    year = {2020}
}

Links

• Paper
• Project page

References

[1] A. Milioto et al., “RangeNet++: Fast and Accurate LiDAR Semantic Segmentation,” in IEEE/RSJ Intl. Conf. on Intelligent Robots and Systems (IROS), 2019.

kitti_scan_unfolding's Issues

Question on pseudo code in function GetRows in algorithm 1

Would you please elaborate on function GetRows? Confused of what the function is going to output. Is it a scalar or a list? Appreiciate it

Image Drawing.

Thanks for your work! Can you share how the image below was drawn?This will be very useful for me！

unfolding.projection give "Cannot find valid image indices for this point cloud and image size. "

I am using your code on the semanticKITTI dataset, and function: unfolding.projection() raise the following error:

"Cannot find valid image indices for this point cloud and image size. "
IndexError: Cannot find valid image indices for this point cloud and image size. Are you sure you entered the correct image size? This function only works with raw KITTI HDL-64 point clouds (no ego motion-corrected data allowed)!

This unfolding.projection() works as expected on the provided "sample_raw.bin".
And I am using ".bin" files from the semanticKITTI dataset.

Are there any changes that I need to make to use your code on the semanticKITTI dataset?
Thanks in advance.

Back projection to pointcloud

Hi Larissa,

First of all, thanks for sharing this code to public. I saw that the projection is so beautiful, it doesn't have a blank rows at the bottom half of the range image. I appreciate that.

I have a question,Is there any option to reproject back to point cloud type? I have tried using another code which uses static vertical resolution, but it is not aligned well with the original pointcloud.

Hope hearing from you soon. Thanks

Question on Figure 2 in the Paper

Hello. I want to ask about the 3D rings you draw in Figure 2. How do you draw figures like this? What software did you use? This figure looks very beautiful. Thanks :)

Any plans for code release?

Hi Triess, I have read your paper and was very impressed. I am currently working on a paper for panoptic segmentation on semanticKITTI and would like to try your scan-based projection approach.
Just wondering is there any plan on releasing the code?
Thanks.