cosense3d.modules.backbone3d package

Submodules

cosense3d.modules.backbone3d.mink_unet module

class cosense3d.modules.backbone3d.mink_unet.MinkUnet(data_info, stride, in_dim, d=3, kernel_size_layer1=3, enc_dim=32, cache_strides=None, floor_height=0, height_compression=None, compression_kernel_size_xy=1, to_dense=False, dist=False, **kwargs)

Bases: BaseModule

QMODE = 1

format_output(res, N)

forward(points: list, **kwargs)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

forward_height_compression(res)

forward_unet(points, **kwargs)

grid_size(stride)

init_weights()

stensor_to_dense(stensor)

to_gpu(gpu_id)

training: bool

valid_coords(stensor)

cosense3d.modules.backbone3d.pillar_bev module

class cosense3d.modules.backbone3d.pillar_bev.PillarBEV(in_channels, layer_nums, layer_strides, downsample_channels, upsample_channels, upsample_strides, voxel_generator, pillar_encoder, bev_shrinker=None, bev_compressor=None, **kwargs)

Bases: BaseModule

format_output(res, N)

forward(points: list, **kwargs)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

to_dense_bev(coor, feat, N)

training: bool

cosense3d.modules.backbone3d.spconv module

class cosense3d.modules.backbone3d.spconv.Spconv(in_channels, out_channels, voxel_generator, voxel_encoder, bev_neck=None, bev_compressor=None, cache_coords=True, cache_strides=[1, 2, 4, 8], **kwargs)

Bases: BaseModule

format_output(out_dict, B)

forward(points: list, **kwargs)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

to_dense(stensor)

training: bool

cosense3d.modules.backbone3d.spconv.post_act_block(in_channels, out_channels, kernel_size, indice_key=None, stride=1, padding=0, conv_type='subm', norm_fn=None)

cosense3d.modules.backbone3d.voxelnet module

class cosense3d.modules.backbone3d.voxelnet.VoxelNet(voxel_generator, voxel_encoder, cml, neck=None, bev_compressor=None, **kwargs)

Bases: BaseModule

forward(points: list, **kwargs)

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

to_dense(coor, feat, N, filter_range=False)

training: bool

Module contents