CNN	convolutional neural networks
FPN	feature pyramid network
DSV	dynamic sparse voxelization
BEV	bird’s-eye view
(x, y, z)	spatial coordinates of the voxels
${\mathrm{f}}_{ }$	corresponding feature map
${\mathrm{d}}_{ }$	inner window voxel ID
N	total voxel window
O	spatial coordinates of all voxels within a subset
V	all voxels within a given window
Q, D	specified subset and voxel
H	size of the feature map
F	size of the convolution kernels
P	size of the padding
S	size of the step
(w, l, h, θ)	length, width, height, and the rotation angle around the z-axis
Ldir	object directional loss
FL	multi-object classification loss
SL	bounding box regression loss
L	total loss