PRBonn / lidar-bonnetal

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

Computation of projected labels during training #59

Closed teja-vsd closed 4 years ago

teja-vsd commented 4 years ago

https://github.com/PRBonn/lidar-bonnetal/blob/423311109bb6a075c8d44f545b624b42a61f8d42/train/tasks/semantic/dataset/kitti/parser.py#L160

Why do you multiply proj_labels with proj_mask when a mask (which is almost similar except for the equality sign) was used during the computation of scan.proj_sem_label. Maybe I am missing something here. I could not get my head around this.

tano297 commented 4 years ago

If you use the single scan case it has no effect, you are free to remove it. This is mostly to ensure that the label only has labeled pixels in the positions where points exist. If you are using semantic-kitti this is always the case, but in other cases where you may be rendering the point clouds from a mesh and also the labels, it may not be like this. Super corner-case which almost never happens.

teja-vsd commented 4 years ago

Thank you so much for your quick reply. This helps!!!