Vertices estimation training

operators.ml3d.vertices_estimation_training(client, in_folder='data_train', in_vertices_folder='data_train_vertices', out_model_path='parameters_model_test', voxel_size=0.02, zero_centering='True', point_names='X,Y,Z', feature_names='', label_names='classification', num_classes=1, label_scales='0.01', learning_rate=1e-05, learning_decay=0.999, num_epochs=2000000, regularization_decay=1e-09, batch_size=5, save_after_epochs=100, backbone_type='MinkUNet14A', head_type_prob='HeadPointwise', criterion_type_prob='L1Sum', hidden_layers=8, max_interpolation_distance=0.75, dist_threshold=0.35, score_threshold=0.4, point_estimation_layers=3, point_estimation_channels=8, criterion_type_point='L1Sum', weight_pred=1.0, weight_prob=1.0, weight_reconstruction=4.0, probabilistic='True', instance_type='x2large')

vertices_estimation_training( client,
in_folder=’data_train’,
in_vertices_folder=’data_train_vertices’,
out_model_path=’parameters_model_test’,
voxel_size=0.02,
zero_centering=’True’,
point_names=’X,Y,Z’,
feature_names=’’,
label_names=’classification’,
num_classes=1,
label_scales=’0.01’,
learning_rate=1e-5,
learning_decay=0.999,
num_epochs=2000000,
regularization_decay=1e-9,
batch_size=5 ,
save_after_epochs=100,
backbone_type=’MinkUNet14A’,
head_type_prob=’HeadPointwise’,
criterion_type_prob=’L1Sum’,
hidden_layers=8,
max_interpolation_distance=0.75,
dist_threshold=0.35,
score_threshold=0.4,
point_estimation_layers=3,
point_estimation_channels=8,
criterion_type_point=’L1Sum’,
weight_pred=1.0,
weight_prob=1.0,
weight_reconstruction=4.0,
probabilistic=’True’,
instance_type=’x2large’ )
Parameters:

  • in_folder – input directory with training data

  • in_vertices_folder – input directory with corresponding vertex data

  • out_model_path – model path

  • voxel_size – voxel size

  • zero_centering – zero centering

  • point_names – point names

  • feature_names – feature names

  • label_names – label names

  • num_classes – number of classes

  • label_scales – label scales

  • learning_rate – learning rate

  • learning_decay – learning rate decay

  • num_epochs – number of epochs

  • regularization_decay – regularization decay

  • batch_size – batch size for training

  • save_after_epochs – save after epochs

  • backbone_type – model type of backbone network

  • head_type_prob – model type of head network

  • criterion_type_prob – model type of criterion

  • hidden_layers – number of hidden layers

  • max_interpolation_distance – maximum distance to interpolate occluded points

  • dist_threshold – distance threshold for non-maximum suppression

  • score_threshold – score threshold for non-maximum suppression

  • point_estimation_layers – number of hidden layers for point estimation

  • point_estimation_channels – number of channels for point estimation

  • criterion_type_point – model type of criterion for point estimation

  • weight_pred – weight for point estimation

  • weight_prob – weight for probability estimation

  • weight_reconstruction – weight for reconstruction estimation

  • probabilistic – probabilistic

  • instance_type – type of cloud instance used for processing