CODEWITHSUNDEEP
labelkittilidar-data
EE_G_RLP
EE_G_RLP

Reputation: 78

How to create BEV labels from 3D-BBox

I want to convert a 3D-BBox of Point Cloud Data to a 2D Bird's Eye Box. I've already created gt_boxes_lidar from gt_boxes_camera with the calibration matrices. My problem is now, that on my created BEV-images (640x640x3) the projected BEV-label doesn't match the object or my approach is false. I tried scaling the 3D-BBox to the Image resolution, which worked fine, but the coordinates for x,y doesn't seem correct, what I've have to do? I'm a little bit clueless.

BEV-image information: res: 640x640x3

cell_size: 0.1m

point_cloud_range: [0, -30, -2.73, 60, 30, 1.27] (x_min, y_min, z_min, x_max, y_max, z_max)

BEV image of sample 000000.bin

Can someone guide me to understand and create succesfully the labels for machine learning?

Update:

Finally with the code from: https://github.com/AlejandroBarrera/birdnet2/blob/5ceed811b289796d7d7420a064ecb079c80801ab/tools/convert_kitti_to_coco_rotation.py#L73 it was possible for me to get matching labels.

Upvotes: 0

Views: 12

Answers (0)

Related Questions