Semanttic output about openpcdet HOT 3 CLOSED

Q1: Inference time is tested on Tesla V100 GPU card.
Q2: The semantic segmentation codes are https://github.com/sshaoshuai/PCDet/blob/master/pcdet/models/rpn/rpn_unet.py#L500
Q3: Yes, you could train on the reduced point clouds for KITTI datasets by removing the get_fov_flag codes https://github.com/sshaoshuai/PCDet/blob/master/pcdet/datasets/kitti/kitti_dataset.py#L414, the performance should be similar.
Q4: What does you mean for static/immovable objects?

from openpcdet.

@sshaoshuai Q4 static objects mean trees building poles

from openpcdet.

If you have GT labels for these objects, you could try to train with them for detecting them. But I think these objects generally are not represented by 3D bounding boxes, and segmentation is more naturally.

from openpcdet.