def get_model(point_cloud: tf.Tensor, features: tf.Tensor, is_training: tf.Variable, num_class: int, bn_decay=None) -> \
[tf.Tensor, tf.Tensor]:
"""
Return a PointNet++ model using additional features as input for the first layer
:param point_cloud: Input points for the model (BxNx3)
:param features: The features for each point (BxNxk)
:param is_training: Flag whether or not the parameters should be trained or not
:param num_class: Number of classes (e.g. 21 for ScanNet)
:param bn_decay: BatchNorm decay
:return: predictions for each point (B x N x num_class)
"""
end_points = {}
l0_xyz = point_cloud
l0_points = features
end_points['l0_xyz'] = l0_xyz
# Layer 1
l1_xyz, l1_points, l1_indices = pointnet_sa_module(l0_xyz,
l0_points,
npoint=1024,
radius=0.1,
nsample=32,
mlp=[32, 32, 64],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer1')
l2_xyz, l2_points, l2_indices = pointnet_sa_module(l1_xyz,
l1_points,
npoint=256,
radius=0.2,
nsample=32,
mlp=[64, 64, 128],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer2')
l3_xyz, l3_points, l3_indices = pointnet_sa_module(l2_xyz,
l2_points,
npoint=64,
radius=0.4,
nsample=32,
mlp=[128, 128, 256],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer3')
l4_xyz, l4_points, l4_indices = pointnet_sa_module(l3_xyz,
l3_points,
npoint=16,
radius=0.8,
nsample=32,
mlp=[256, 256, 512],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer4')
# Feature Propagation layers
l3_points = pointnet_fp_module(l3_xyz,
l4_xyz,
l3_points,
l4_points, [256, 256],
is_training,
bn_decay,
scope='fa_layer1')
l2_points = pointnet_fp_module(l2_xyz,
l3_xyz,
l2_points,
l3_points, [256, 256],
is_training,
bn_decay,
scope='fa_layer2')
l1_points = pointnet_fp_module(l1_xyz,
l2_xyz,
l1_points,
l2_points, [256, 128],
is_training,
bn_decay,
scope='fa_layer3')
l0_points = pointnet_fp_module(l0_xyz,
l1_xyz,
l0_points,
l1_points, [128, 128, 128],
is_training,
bn_decay,
scope='fa_layer4')
# Full connected layers
net = tf_util.conv1d(l0_points,
128,
1,
padding='VALID',
bn=True,
is_training=is_training,
scope='fc1',
bn_decay=bn_decay)
end_points['feats'] = net
net = tf_util.dropout(net,
keep_prob=0.5,
is_training=is_training,
scope='dp1')
net = tf_util.conv1d(net,
num_class,
1,
padding='VALID',
activation_fn=None,
scope='fc2')
return net, end_points
def call(self, inputs, **kwargs):
l0_xyz = inputs
l0_points = None
l1_xyz, l1_points, l1_indices = self.l1([l0_xyz, tf.zeros([0])])
print("l1_points shape: ", l1_points.shape)
l2_xyz, l2_points, l2_indices = self.l2([l1_xyz, l1_points])
print("l2_points shape: ", l2_points.shape)
l3_xyz, l3_points, l3_indices = self.l3([l2_xyz, l2_points])
print("l3_points shape: ", l3_points.shape)
l4_xyz, l4_points, l4_indices = self.l4([l3_xyz, l3_points])
print("l4_points shape: ", l4_points.shape)
# Feature Propagation layers
l3_points = pointnet_fp_module(l3_xyz,
l4_xyz,
l3_points,
l4_points, [256, 256],
self.is_training,
self.bn_decay,
scope='fa_layer1')
l2_points = pointnet_fp_module(l2_xyz,
l3_xyz,
l2_points,
l3_points, [256, 256],
self.is_training,
self.bn_decay,
scope='fa_layer2')
l1_points = pointnet_fp_module(l1_xyz,
l2_xyz,
l1_points,
l2_points, [256, 128],
self.is_training,
self.bn_decay,
scope='fa_layer3')
l0_points = pointnet_fp_module(l0_xyz,
l1_xyz,
l0_points,
l1_points, [128, 128, 128],
self.is_training,
self.bn_decay,
scope='fa_layer4')
# FC layers
net = tf_util.conv1d(l0_points,
128,
1,
padding='VALID',
bn=True,
is_training=self.is_training,
scope='fc1',
bn_decay=self.bn_decay)
net = tf_util.dropout(net,
keep_prob=0.5,
is_training=self.is_training,
scope='dp1')
out = tf_util.conv1d(net,
self.num_class,
1,
padding='VALID',
activation_fn=None,
scope='fc2')
return out
def get_model(point_cloud, is_training, num_class, bn_decay=None):
""" Semantic segmentation PointNet, input is BxNx3, output Bxnum_class """
batch_size = point_cloud.get_shape()[0].value
num_point = point_cloud.get_shape()[1].value
end_points = {}
l0_xyz = point_cloud
l0_points = None
end_points['l0_xyz'] = l0_xyz
# Layer 1
l1_xyz, l1_points, l1_indices = pointnet_sa_module(l0_xyz,
l0_points,
npoint=1024,
radius=0.1,
nsample=32,
mlp=[32, 32, 64],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer1')
l2_xyz, l2_points, l2_indices = pointnet_sa_module(l1_xyz,
l1_points,
npoint=256,
radius=0.2,
nsample=32,
mlp=[64, 64, 128],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer2')
l3_xyz, l3_points, l3_indices = pointnet_sa_module(l2_xyz,
l2_points,
npoint=64,
radius=0.4,
nsample=32,
mlp=[128, 128, 256],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer3')
l4_xyz, l4_points, l4_indices = pointnet_sa_module(l3_xyz,
l3_points,
npoint=16,
radius=0.8,
nsample=32,
mlp=[256, 256, 512],
mlp2=None,
group_all=False,
is_training=is_training,
bn_decay=bn_decay,
scope='layer4')
# Feature Propagation layers
l3_points = pointnet_fp_module(l3_xyz,
l4_xyz,
l3_points,
l4_points, [256, 256],
is_training,
bn_decay,
scope='fa_layer1')
l2_points = pointnet_fp_module(l2_xyz,
l3_xyz,
l2_points,
l3_points, [256, 256],
is_training,
bn_decay,
scope='fa_layer2')
l1_points = pointnet_fp_module(l1_xyz,
l2_xyz,
l1_points,
l2_points, [256, 128],
is_training,
bn_decay,
scope='fa_layer3')
l0_points = pointnet_fp_module(l0_xyz,
l1_xyz,
l0_points,
l1_points, [128, 128, 128],
is_training,
bn_decay,
scope='fa_layer4')
# FC layers
net = tf_util.conv1d(l0_points,
128,
1,
padding='VALID',
bn=True,
is_training=is_training,
scope='fc1',
bn_decay=bn_decay)
end_points['feats'] = net
net = tf_util.dropout(net,
keep_prob=0.5,
is_training=is_training,
scope='dp1')
net = tf_util.conv1d(net,
num_class,
1,
padding='VALID',
activation_fn=None,
scope='fc2')
return net, end_points