Hi guys, I’m sorry for the inconvenience. I would like to understand how to use open3D codes with deep learning (tensorflow). can anybody help me? An example below:
def fpfh(vector):
lf = []
x,bat=vector
for LF in range(bat):
Cloud=x[LF]
pcd = o3d.geometry.PointCloud()
pcd.points = o3d.utility.Vector3dVector(Cloud)
#np.asarray(pcd.points)
voxel_size = 0.05 # means 5cm for the dataset
radius_normal = voxel_size * 2
pcd.estimate_normals(o3d.geometry.KDTreeSearchParamHybrid(radius=radius_normal, max_nn=30))
radius_feature = voxel_size * 5
pcd_fpfh = o3d.pipelines.registration.compute_fpfh_feature(pcd,o3d.geometry.KDTreeSearchParamHybrid(radius=radius_feature, max_nn=100))
localfeature = pcd_fpfh.data
localfeature = StandardScaler().fit_transform(localfeature)
localfeature = localfeature.T
lf.append(localfeature)
xlf=np.array(lf)
return tf.convert_to_tensor(xlf)
def conv_bn(x, filters):
x = Conv1D(filters, kernel_size=1, kernel_regularizer=l2(1e-4), use_bias = True, bias_regularizer=l2(1e-4), padding=“valid”)(x)
x = BatchNormalization(momentum=0.0)(x)
return Activation(“relu”)(x)
def dense_bn(x, filters):
x = Dense(filters, kernel_regularizer=l2(1e-4), use_bias = True, bias_regularizer=l2(1e-4))(x)
x = BatchNormalization(momentum=0.0)(x)
return Activation(“relu”)(x)
def tnet(inputs, num_features):
bias = keras.initializers.Constant(np.eye(num_features).flatten())
x = conv_bn(inputs, 32)
x = conv_bn(x, 64)
x = conv_bn(x, 512)
x = GlobalMaxPooling1D()(x)
x = dense_bn(x, 256)
x = dense_bn(x, 128)
x = Dense(
num_features * num_features,
kernel_regularizer=l2(1e-4),
use_bias = True,
kernel_initializer=“GlorotNormal”,
bias_initializer=bias,
activity_regularizer=l2(1e-4)
)(x)
feat_T = Reshape((num_features, num_features))(x)
# Apply affine transformation to input features
return Dot(axes=(2, 1))([inputs, feat_T])
input_pc = Input(shape=(num_points, 3),name=‘entrada’)
x = tnet(input_pc, 3) # T-net layers.
bat = x.shape[0]
xlf = tf.keras.layers.Lambda(function=fpfh)([x,bat])