ocnn.models

`LeNet`	Octree-based LeNet for classification.
`ResNet`	Octree-based ResNet for classification.
`SegNet`	Octree-based SegNet for segmentation.
`UNet`	Octree-based UNet for segmentation.
`HRNet`	Octree-based HRNet for classification and segmentation.
`AutoEncoder`	Octree-based AutoEncoder for shape encoding and decoding.
`OUNet`
`Image2Shape`	Octree-based AutoEncoder for shape encoding and decoding.

class LeNet(in_channels: int, out_channels: int, stages: int, nempty: bool = False)[source]

Octree-based LeNet for classification.

forward(data: Tensor, octree: Octree, depth: int)[source]

class ResNet(in_channels: int, out_channels: int, resblock_num: int, stages: int, nempty: bool = False, dropout: float = 0.5)[source]

Octree-based ResNet for classification.

forward(data: Tensor, octree: Octree, depth: int)[source]

class SegNet(in_channels: int, out_channels: int, stages: int, interp: str = 'linear', nempty: bool = False, **kwargs)[source]

Octree-based SegNet for segmentation.

forward(data: Tensor, octree: Octree, depth: int, query_pts: Tensor)[source]

class UNet(in_channels: int, out_channels: int, interp: str = 'linear', nempty: bool = False, **kwargs)[source]

Octree-based UNet for segmentation.

config_network()[source]: Configure the network channels and Resblock numbers.

unet_encoder(data: Tensor, octree: Octree, depth: int)[source]: The encoder of the U-Net.

unet_decoder(convd: Dict[int, Tensor], octree: Octree, depth: int)[source]: The decoder of the U-Net.

forward(data: Tensor, octree: Octree, depth: int, query_pts: Tensor)[source]

class HRNet(in_channels: int, out_channels: int, stages: int = 3, interp: str = 'linear', nempty: bool = False)[source]

Octree-based HRNet for classification and segmentation.

forward(data: Tensor, octree: Octree, depth: int)[source]

class AutoEncoder(channel_in: int, channel_out: int, depth: int, full_depth: int = 2, feature: str = 'ND')[source]

Octree-based AutoEncoder for shape encoding and decoding.

Parameters:

channel_in (int) – The channel of the input signal.
channel_out (int) – The channel of the output signal.
depth (int) – The depth of the octree.
full_depth (int) – The full depth of the octree.
feature (str) – The feature type of the input signal. For details of this argument, please refer to ocnn.modules.InputFeature.

encoder(octree: Octree)[source]: The encoder network of the AutoEncoder.

decoder(shape_code: Tensor, octree: Octree, update_octree: bool = False)[source]: The decoder network of the AutoEncoder.

decode_code(shape_code: Tensor)[source]

Decodes the shape code to an output octree.

Parameters:: shape_code (torch.Tensor) – The shape code for decoding.

init_octree(shape_code: Tensor)[source]

Initialize a full octree for decoding.

Parameters:: shape_code (torch.Tensor) – The shape code for decoding, used to get the batch_size and device to initialize the output octree.

forward(octree: Octree, update_octree: bool)[source]

class OUNet(channel_in: int, channel_out: int, depth: int, full_depth: int = 2, feature: str = 'ND')[source]

encoder(octree)[source]: The encoder network for extracting heirarchy features.

decoder(convs: dict, octree_in: Octree, octree_out: Octree, update_octree: bool = False)[source]: The decoder network for decode the octree.

init_octree(octree_in: Octree)[source]: Initialize a full octree for decoding.

forward(octree_in, octree_out=None, update_octree: bool = False)[source]

class Image2Shape(channel_out: int, depth: int, full_depth: int = 2, code_channel: int = 32)[source]