search

TRAILab / CaDDN

Categorical Depth Distribution Network for Monocular 3D Object Detection (CVPR 2021 Oral)
Apache License 2.0
359 stars 62 forks source link

About the Coordinate Space #46

Closed czy341181 closed 3 years ago

czy341181 commented 3 years ago

When building the voxel, it is constructed in the point cloud coordinate system. So the output heatmap map is also in the point cloud coordinate system (so during training, the 3D box is converted to the lidar coordinate system as the label). But the final 3D Box in the inference stage is based on lidar coordinate, so it still needs to be converted to camera coodinate?

Do I understand correctly?

czy341181 commented 3 years ago

Sorry, In my description。 point cloud coordinate system == lidar coordinate system [2, -30.08, -3.0, 46.8, 30.08, 1.0]

czy341181 commented 3 years ago

I think my understanding is right. I see the code

https://github.com/TRAILab/CaDDN/blob/5a96b37f16b3c29dd2509507b1cdfdff5d53c558/pcdet/datasets/kitti/kitti_dataset.py#L350

Is it right? Thank you very much.

codyreading commented 3 years ago

Yes that is correct. I inherited most of the code from OpenPCDet which does all of its representations in the LIDAR coordinate system. The same system is used here.