mmdetection3d icon indicating copy to clipboard operation
mmdetection3d copied to clipboard

waymo convert about CALIB file in kitti format!

Open xlb8 opened this issue 2 years ago • 1 comments

the parameters in camera.intrinsic [0:4] is the projection parameters toward to the waymo camera coordinate, and the tr_velo_to_cam(after transform) contains the transform from waymo camera coor to kitti's (T_front_cam_to_ref),but the projection parameter didnt,you just saved the 4 parameter without any transformation, i wonder if its a issue or some transformation i dont understand?

xlb8 avatar Jul 25 '22 10:07 xlb8

Could you please describe this issue more clearly with some references to specific codes?

Tai-Wang avatar Aug 03 '22 06:08 Tai-Wang

I wonder in save_calib() in tools/dataset_converters/waymo_converter.py, why T_vehicle_to_cam is renamed as Tr_velo_to_cam? I think the vehicle frame and the LiDAR frame are not the same thing in Waymo. T_vehicle_to_cam is derived from camera.extrinsic, which is indeed the transformation from vehicle frame to camera frame. But how can it be renamed and used as the transformation from LiDAR to camera? Do I misunderstand anything?

xyupeng avatar Nov 08 '22 14:11 xyupeng

I wonder in save_calib() in tools/dataset_converters/waymo_converter.py, why T_vehicle_to_cam is renamed as Tr_velo_to_cam? I think the vehicle frame and the LiDAR frame are not the same thing in Waymo. T_vehicle_to_cam is derived from camera.extrinsic, which is indeed the transformation from vehicle frame to camera frame. But how can it be renamed and used as the transformation from LiDAR to camera? Do I misunderstand anything?

point_in_cam = point_in_velo@Tr_velo_to_cam, i think the "cam" in Tr_velo_to_cam is just a virtual coordinate in kitti, it is just a bridge, as long as i can get the right point in camera coordinate, and it can be project to the ref image,it works. no 1 to 1 definition in this process, just pay attention to the results of transformation.

in addition, the key you need to know is that the points extracted have been place to the vehicle frame, so there's no "LiDAR coordinate".Point Cloud Conversion and Visualization

xlb8 avatar Nov 08 '22 15:11 xlb8

Thank you so much for the clarification. I think I missed the point cloud extraction part which has already transformed the point cloud to the vehicle frame.

xyupeng avatar Nov 09 '22 00:11 xyupeng