lingvo.tasks.car.pointnet module

PointNet architecture.

PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation

https://arxiv.org/abs/1612.00593

class lingvo.tasks.car.pointnet.PointNet

Bases: lingvo.tasks.car.builder_lib.ModelBuilderBase

Builder for PointNet layers.

_ConcatWithBranch(name, tile, *subs)

Concat x with subs(x).

_ConcatWithOnehot(name)

Make an onehot from shape category id and concat with point features.

See original Pointnet paper for justification.

Parameters

name – Name of the layer.

Returns

Params for a layer that updates the NestedMap features value with the input features + a one hot encoding of the category id.

_TNet(name, idims)

Classifier(name='pointnet', keep_prob=0.7, input_dims=3, feature_dims=256)

PointNet architecture for classification.

Segmentation(name='pointnet_segmentation')

PointNet archetecture for segmentation.

SegmentationShapeNet(name='pointnet_shapenet', keep_prob=0.8)

The modified network for part segmentation in ShapeNet.

class lingvo.tasks.car.pointnet.PointNetPP

Bases: lingvo.tasks.car.builder_lib.ModelBuilderBase

Builder for PointNet++ Model.

_SetAbstractionWithMLPMax(name, mlp_dims, num_samples, group_size, ball_radius, sample_neighbors_uniformly=True)

_ModelNet40Featurizer(input_dims)

Classifier(name='pointnetpp', input_dims=3, feature_dims=256, keep_prob=0.8, num_points=1024)

PointNet++ architecture for modelnet40 classification.