def pc_sampler(mesh_file, n_samples, save_file=None, dtype=np.float32):
in_mesh = Mesh(file_name=mesh_file)
in_mesh = cleaning.clean_mesh(in_mesh)
in_mesh.swap_axes_of_vertices([0, 2, 1])
ss_points, _ = in_mesh.sample_faces(n_samples)
pc = Point_Cloud(points=ss_points.astype(dtype))
pc.center_in_unit_sphere()
pc, _ = pc.lex_sort()
if save_file is not None:
pc.save_as_ply(save_file)
return pc
def main_Mesh():
off_file = '/Users/optas/DATA/Shapes/Model_Net_10/OFF_Original/bathtub/train/bathtub_0001.off'
in_mesh = Mesh(off_file=off_file)
in_mesh.center_in_unit_sphere()
cleaning.clean_mesh(in_mesh, level=3, verbose=True)
in_lb = Laplace_Beltrami(in_mesh)
n_cc, node_labels = in_mesh.connected_components()
parts_id = Graph.largest_connected_components_at_thres(node_labels, 1)
print 'Number of connected components = %d.' % (n_cc)
percent_of_eigs = 1
min_eigs = None
max_eigs = None
min_vertices = None
time_horizon = 10
area_type = 'barycentric'
v_color = ns.hks_of_component_spectra(in_mesh, in_lb, area_type, percent_of_eigs, \
time_horizon, min_vertices, min_eigs, max_eigs)[0]
in_mesh.plot(vertex_function=v_color)
def pc_uniform_sampler(mesh_file, n_samples, swap_y_z=True, save_file=None, dtype=np.float32, out_folder=None):
''' Given a mesh, it computes a point-cloud that is uniformly sampled
from its area elements.
'''
in_mesh = Mesh(file_name=mesh_file)
if swap_y_z:
in_mesh.swap_axes_of_vertices([0, 2, 1])
in_mesh = cleaning.clean_mesh(in_mesh)
ss_points, _ = in_mesh.sample_faces(n_samples)
pc = Point_Cloud(points=ss_points.astype(dtype))
pc.center_in_unit_sphere()
pc, _ = pc.lex_sort()
if save_file is not None:
pc.save_as_ply(save_file)
return pc
def pc_sampler(mesh_file,
n_samples,
save_file=None,
rotate=False,
dtype=np.float32):
category = file_to_category(mesh_file)
if rotate:
rotate_deg = rotation_angles[category]
in_mesh = Mesh(file_name=mesh_file)
in_mesh = cleaning.clean_mesh(in_mesh)
ss_points, _ = in_mesh.sample_faces(n_samples)
pc = Point_Cloud(points=ss_points.astype(dtype))
if rotate:
pc.rotate_z_axis_by_degrees(rotate_deg)
pc.center_in_unit_sphere()
pc, _ = pc.lex_sort()
if save_file is not None:
pc.save_as_ply(save_file)
return pc
if laplacian_type == 'comb':
L = -A + D
elif laplacian_type == 'norm':
total_weight = (1 / np.sqrt(total_weight)).squeeze()
Dn = sparse.spdiags(total_weight, 0, n, n)
L = Dn.dot(-A + D).dot(Dn)
elif laplacian_type == 'sign':
L = A + D
else:
raise ValueError('Please provide a valid argument for the type of laplacian.')
return L
if __name__ == '__main__':
from geo_tool.solids import mesh_cleaning as cleaning
from geo_tool.solids.mesh import Mesh
off_file = '/Users/t_achlp/Documents/DATA/ModelNet10/OFF_Original/bathtub/train/bathtub_0001.off'
in_mesh = Mesh(off_file)
in_mesh.center_in_unit_sphere()
cleaning.clean_mesh(in_mesh, level=3, verbose=False)
cloud_points, face_ids = in_mesh.sample_faces(2000)
from scipy import spatial
tree = spatial.KDTree(cloud_points, leafsize=100)
weights, neighbors = tree.query(cloud_points, 10)
weights = weights[:, 1:]
neighbors = neighbors[:, 1:]
weights = np.exp(- weights**2 / (2 * np.median(weights)))
A = Graph.knn_to_adjacency(neighbors, weights)