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

Using the code on less number of beams #9

Closed izzat1 closed 5 years ago

izzat1 commented 5 years ago

Great work. Thanks for sharing the dataset and code. I tried you code on a 40 beam lidar and did not get any useful result using your trained weights (darknet53-1024). I made changes to the laserscan.py (fov, and H to match my lidar) and also adjusted the height. Any other change I need to make? I created a .bin file from my data and used it as input to your infer.py code.

tano297 commented 5 years ago

This is related to https://github.com/PRBonn/lidar-bonnetal/issues/4

izzat1 commented 5 years ago

Yes I reviewed your response and tried to change the data to match Kitti. I expected some performance degradation but not as much as I am seeing. Most of the points are classified as the last class. Maybe some form of domain adaptation is needed to make things work well.