search

PRBonn / lidar-bonnetal

Semantic and Instance Segmentation of LiDAR point clouds for autonomous driving
http://semantic-kitti.org
MIT License
959 stars 206 forks source link

Explanation on how to deal with this repo #57

Closed FrancescoMandru closed 4 years ago

FrancescoMandru commented 4 years ago

Hello, I would like to ask you if someone can explain to me how to work with this repo since there are a lot of informations and I don't know what to do.

I would like to get the RangeNet++ trained, pass my 3D KITTI LiDAR point cloud and get for each point the corresponding label.

I don't know how to do that, thank you in advance

jbehley commented 4 years ago

  1. For RangeNet++, you need to download the pre-trained model: http://www.ipb.uni-bonn.de/html/projects/bonnetal/lidar/semantic/models/darknet53.tar.gz As the text says: "To enable kNN post-processing, just change the boolean value to True in the arch_cfg.yaml file parameter, inside the model directory."

  2. You want to infer the labels, then you have to look here: https://github.com/PRBonn/lidar-bonnetal/tree/master/train/tasks/semantic#inference, where you can give it the dataset (Your dataset in the folder structure of KITTI), the folder to put the predictions (output folder), and the model (which you downloaded in 1.).

  3. The labels outputed are in the SemanticKITTI format, which stores for each point a label, which you can read with numpy.fromfile("/path/to/predictions/00000.label", dtype=numpy.uint32).