spheres_README.md

Dynamic Plane + Spheres Convolutional Occupancy Network

Table of Contents

• Sphere Encoding
• Sphere Radius-Map Encoding
• Sphere Voxel Encoding
• Multi-Learning Spheres Encoding
• Multi-Const Spheres Encoding
• Point_Plane_Sphere_Height_map Encoding

Sphere Encoding

Sphere	Sphere Discretisation

python train_point_plane.py sphere.yaml model_name

Use sphere_net in encoder and sphere in decoder. Put one_sphere in interpolation in the decoder_kwargs. You may change the discretisation of sphere (number of meridians/latitudes) by changing the parameter of latitudes/meridians in encoder_kwargs and decoder_kwargs

Sphere Radius-Map Encoding

python train_point_plane.py sphere_height.yaml model_name

Use sphere_height_net in encoder and sphere in decoder. Put one_sphere in interpolation in the decoder_kwargs. You may change the discretisation of sphere (number of meridians/latitudes) by changing the parameter of latitudes/meridians in encoder_kwargs and decoder_kwargs

Sphere Voxel Encoding

python train_point_plane.py multi_voxel.yaml model_name

Use multi_voxels_sphere in encoder and sphere in decoder. Put multi_voxel_sphere in interpolation in the decoder_kwargs. You may change the number of "depth-layers" (in the radious dimension) by changing the parameter of rad_dim in encoder_kwargs and decoder_kwargs

Multi-Learning Spheres Encoding

python train_point_plane.py multi_sphere.yaml model_name

Use multi_sphere_net in encoder and sphere in decoder. Put multi_spheres in interpolation in the decoder_kwargs. You may change the number of learned spheres by changing the parameter of n_channels in encoder_kwargs

Multi-Const Spheres Encoding

python train_point_plane.py multi_const_spheres.yaml model_name

Use const_multi_spheres in encoder and sphere in decoder. Put multi_spheres in interpolation in the decoder_kwargs. You may change the number of constant-positioned spheres by changing the parameter of n_channels in encoder_kwargs and then change the position of spheres-centers manually in the class const_multi_spheres

Point_Plane_Sphere_Height_map Encoding

python train_point_plane.py sphere_plane_height.yaml model_name

Use sphere_height_plane_net in encoder and cbatchnorm in decoder. You may change the number of learned planes by changing the parameter of n_channels in encoder_kwargs and then change the position of spheres-centers manually in the class const_multi_spheres