Differentiable voxelization operator using Pytorch

Compatibale with OpenPCDet (https://github.com/open-mmlab/OpenPCDet)

• Voxelization operator

  from voxel_ops import Voxelization
  
  voxel_module = Voxelization(voxel_size=[0.16, 0.16, 4],
                              point_cloud_range=[0, -39.68, -3, 69.12, 39.68, 1],
                              max_num_points=32, # number of points per voxel
                              max_voxels=40000,
                              deterministic=True,  # differentiable only if deterministic==True
                              )

• Work with OpenPCDet

  - NAME: differentiable_voxelize
  	RETAIN_GRAPH: False
  	END_TO_END_MODE: False
  	DETERMINISTIC: True
  	VOXEL_SIZE: [0.16, 0.16, 4]
  	MAX_POINTS_PER_VOXEL: 32
  	MAX_NUMBER_OF_VOXELS: {
  		'train': 16000,
  		'test': 40000
            }

a. Install pytorch properly.

b. Clone this repository.

git clone [email protected]:Uzukidd/voxelization.git

c. Install voxel-ops from source.

cd ./voxelization
pip install -e .

See ./test/ for more details