hm2092

From my understanding, 2 is the number of anchors per position. One of them is oriented at 0 and other one at pi/2. w and l are the dimensions of...

You can refer this repo: https://github.com/skyhehe123/VoxelNet-pytorch/blob/bfd7be1ca6e5c5b39eece8a4c5fad3ad7b1f7c74/utils.py#L145

The above code uses KITTI dataset in which the labels(x, y, z coordinates for bounding box centre) are given in camera coordinates. Thats why it is converted first to lidar...

I am working on a pytorch implementation of the same and i found that the network for car detects also pedestrians and cyclists, even though i use anchor boxes in...