def reconstruct_npz(inname, outname):
"""
Recontruct a 3D shape by deforming a template
:param inname: input path
:return: None (but save reconstruction)
"""
if os.path.exists(outname):
return
with np.load(inname) as npl:
V, F = npl['V'], npl['F']
V = pca_whiten(V)
max_axis = np.argmax((np.max(V, axis=0) - np.min(V, axis=0)))
V = V[:, np.roll(np.arange(3), 1 - max_axis)] # 1 means Y
V *= 1.7
assert (np.max(V, axis=0) - np.min(V, axis=0))[1] > 1.69
while V.shape[0] < 1e4:
eV, eF = p2e(V), p2e(F)
NV, NF = igl.eigen.MatrixXd(), igl.eigen.MatrixXi()
igl.upsample(eV, eF, NV, NF)
V, F = e2p(NV), e2p(NF)
input = trimesh.Trimesh(vertices=V, faces=F, process=False)
scalefactor = 1.0
if global_variables.opt.scale:
input, scalefactor = scale(
input, global_variables.mesh_ref_LR
) #scale input to have the same volume as mesh_ref_LR
if global_variables.opt.clean:
input = clean(input) #remove points that doesn't belong to any edges
test_orientation(input)
inp_V = input.vertices
if inp_V.shape[0] > 1e5:
inp_V = inp_V[
np.random.choice(inp_V.shape[0], int(1e5), replace=False), :]
final_points, final_loss = run(inp_V, scalefactor)
npz_path = os.path.dirname(outname)
if not os.path.exists(npz_path): os.makedirs(npz_path)
np.savez(outname, V=final_points, l=final_loss)
def query(self, queries):
"""Returns numpy array of SDF values for each point in queries"""
queryV = iglhelpers.p2e(queries)
S = igl.eigen.MatrixXd()
B = igl.eigen.MatrixXd()
I = igl.eigen.MatrixXi()
C = igl.eigen.MatrixXd()
N = igl.eigen.MatrixXd()
if self._precomputed and self._signType == igl.SIGNED_DISTANCE_TYPE_FAST_WINDING_NUMBER:
# generate samples from precomputed bvh's
print("[INFO] Generating SDFs")
igl.signed_distance_fast_winding_number(queryV, self._V, self._F,
self._tree, self._fwn_bvh,
S)
print("[INFO] SDFs done")
else:
igl.signed_distance(queryV, self._V, self._F, self._signType, S, I,
C, N)
return iglhelpers.e2p(S)
class Geom():
def __init__(self, s=None):
self.V = igl.eigen.MatrixXd()
self.F = igl.eigen.MatrixXi()
if s is not None:
igl.read_triangle_mesh(s, self.V, self.F)
# Picking Largest Component
from scipy import stats
for path in scans:
Inmodel = Geom(path)
C = igl.eigen.MatrixXi()
igl.facet_components(Inmodel.F, C)
C = (e2p(C)).flatten()
modeC, mode_count = stats.mode(C)
if mode_count == Inmodel.F.rows():
print(f"Already Fit {path[-15:-12]}")
igl.write_triangle_mesh(path[:-12] + '_single.obj', Inmodel.V,
Inmodel.F)
else:
Fid = p2e(np.where(C == modeC)[0])
F = igl.slice(Inmodel.F, Fid, 1)
Outmodel = Geom()
I, J = igl.eigen.MatrixXi(), igl.eigen.MatrixXi()
igl.remove_unreferenced(Inmodel.V, F, Outmodel.V, Outmodel.F, I, J)
igl.write_triangle_mesh(path[:-12] + '_single.obj', Outmodel.V,
Outmodel.F)
print(f'Written { path[-15:-12]}')
def setup_deformation_transfer(source, target, use_normals=False):
rows = np.zeros(3 * target.v.shape[0])
cols = np.zeros(3 * target.v.shape[0])
coeffs_v = np.zeros(3 * target.v.shape[0])
coeffs_n = np.zeros(3 * target.v.shape[0])
print("Computing nearest vertices")
P = p2e(target.v)
V = p2e(source.v)
F = p2e(source.f)
sqrD = igl.eigen.MatrixXd()
nearest_faces = igl.eigen.MatrixXi()
nearest_vertices = igl.eigen.MatrixXd()
igl.point_mesh_squared_distance(P, V, F, sqrD, nearest_faces,
nearest_vertices)
print("Computing barycentric coordinates")
coeffs_v = igl.eigen.MatrixXd()
Va, Vb, Vc = igl.eigen.MatrixXd(), igl.eigen.MatrixXd(
), igl.eigen.MatrixXd()
F2 = igl.eigen.MatrixXi()
xyz = p2e(np.array([0, 1, 2]))
igl.slice(F, nearest_faces, xyz, F2)
igl.slice(V, F2.col(0), xyz, Va)
igl.slice(V, F2.col(1), xyz, Vb)
igl.slice(V, F2.col(2), xyz, Vc)
igl.barycentric_coordinates(nearest_vertices, Va, Vb, Vc, coeffs_v)
nearest_faces = e2p(nearest_faces)
coeffs_v = e2p(coeffs_v).ravel()
rows = np.array([i for i in range(target.v.shape[0]) for _ in range(3)])
cols = source.f[nearest_faces].ravel()
"""
nearest_faces, nearest_parts, nearest_vertices = source.compute_aabb_tree().nearest(target.v, True)
nearest_faces = nearest_faces.ravel().astype(np.int64)
nearest_parts = nearest_parts.ravel().astype(np.int64)
nearest_vertices = nearest_vertices.ravel()
for i in range(target.v.shape[0]):
# Closest triangle index
f_id = nearest_faces[i]
# Closest triangle vertex ids
nearest_f = source.f[f_id]
# Closest surface point
nearest_v = nearest_vertices[3 * i:3 * i + 3]
# Distance vector to the closest surface point
dist_vec = target.v[i] - nearest_v
rows[3 * i:3 * i + 3] = i * np.ones(3)
cols[3 * i:3 * i + 3] = nearest_f
n_id = nearest_parts[i]
if n_id == 0:
# Closest surface point in triangle
A = np.vstack((source.v[nearest_f])).T
coeffs_v[3 * i:3 * i + 3] = np.linalg.lstsq(A, nearest_v)[0]
elif n_id > 0 and n_id <= 3:
# Closest surface point on edge
A = np.vstack((source.v[nearest_f[n_id - 1]], source.v[nearest_f[n_id % 3]])).T
tmp_coeffs = np.linalg.lstsq(A, target.v[i])[0]
coeffs_v[3 * i + n_id - 1] = tmp_coeffs[0]
coeffs_v[3 * i + n_id % 3] = tmp_coeffs[1]
else:
# Closest surface point a vertex
coeffs_v[3 * i + n_id - 4] = 1.0
"""
# if use_normals:
# A = np.vstack((vn[nearest_f])).T
# coeffs_n[3 * i:3 * i + 3] = np.linalg.lstsq(A, dist_vec)[0]
#coeffs = np.hstack((coeffs_v, coeffs_n))
#rows = np.hstack((rows, rows))
#cols = np.hstack((cols, source.v.shape[0] + cols))
matrix = sp.csc_matrix((coeffs_v, (rows, cols)),
shape=(target.v.shape[0], source.v.shape[0]))
return matrix
import numpy as np
import pygraphcut
import sys, os
sys.path.append(os.path.expanduser('~/Workspace/libigl/python'))
import pyigl as igl
from iglhelpers import e2p, p2e
with np.load(
'/home/zhongshi/data/sig17_seg_benchmark/full_dump0/shrec/10.off.npz'
) as npl:
print(dict(npl).keys())
V, F = npl['V'], npl['F']
new_label = np.asarray(pygraphcut.sdf_values(V, F))
import pdb
pdb.set_trace()
vw = igl.glfw.Viewer()
vw.data().set_mesh(p2e(V), p2e(F))
vw.data().set_colors(p2e(new_label))
vw.launch()
print(new_label)
import pymmgs
import numpy as np
import sys, os
sys.path.insert(0, os.path.expanduser('~/Workspace/libigl/python'))
import pyigl as igl
import pyigl.eigen as Eigen
from iglhelpers import p2e, e2p
V = igl.eigen.MatrixXd()
F = igl.eigen.MatrixXi()
igl.read_triangle_mesh('/Users/zhongshi/1399_standing_clap_000010.obj', V, F)
# SV, SVI, SVJ, SF = igl.eigen.MatrixXd(), igl.eigen.MatrixXi(), igl.eigen.MatrixXi(), igl.eigen.MatrixXi()
# igl.remove_duplicate_vertices(V,F,1e-10, SV, SVI, SVJ, SF)
M = igl.eigen.MatrixXd()
igl.doublearea(V, F, M)
M = e2p(M).flatten()
# F0 = e2p(F)[np.where(M > 1e-9)[0],:]
# npl = np.load('/Users/zhongshi/1006_jump_from_wall_000003.obj.npz')
V0, F0 = e2p(V), e2p(F)
# igl.writeOBJ('1399.obj',V, F)
V, F = pymmgs.MMGS(V0, F0, 0.005)
vw = igl.glfw.Viewer()
vw.data().set_mesh(p2e(V), p2e(F))
vw.launch()
def march(self, pts, sdf):
res = round(pts.shape[0]**(1 / 3))
v = iglhelpers.p2e(pts)
s = iglhelpers.p2e(sdf)
igl.copyleft.marching_cubes(s, v, res, res, res, self._rV, self._rF)
self.hasMarched = True
def main():
global verts_sample
initial_verts, initial_faces = get_initial_verts_and_faces()
numpy_base_verts = e2p(initial_verts).flatten()
start = time.time()
num_samples = 250
numpy_verts_sample = load_samples(num_samples)
numpy_displacements_sample = numpy_verts_sample - initial_verts
num_verts = len(numpy_verts_sample[0])
print(num_verts)
print('Loading...')
verts_sample = [p2e(m) for m in numpy_verts_sample]
displacements_sample = [p2e(m) for m in numpy_displacements_sample]
print("Took:", time.time() - start)
use_pca = False
if use_pca:
### PCA Version
print("Doing PCA...")
train_size = num_samples
test_size = num_samples
test_data = numpy_displacements_sample[:test_size] * 1.0
test_data_eigen = verts_sample[:test_size]
numpy.random.shuffle(numpy_displacements_sample)
# train_data = numpy_verts_sample[test_size:test_size+train_size]
train_data = numpy_displacements_sample[0:train_size]
pca = PCA(n_components=3)
pca.fit(train_data.reshape((train_size, 3 * num_verts)))
# def encode(q):
# return pca.transform(numpy.array([q.flatten() - numpy_base_verts]))[0]
# def decode(z):
# return (numpy_base_verts + pca.inverse_transform(numpy.array([z]))[0]).reshape((num_verts, 3))
# print(numpy.equal(test_data[0].flatten().reshape((len(test_data[0]),3)), test_data[0]))
# print(encode(test_data[0]))
test_data_encoded = pca.transform(
test_data.reshape(test_size, 3 * num_verts))
test_data_decoded = (numpy_base_verts +
pca.inverse_transform(test_data_encoded)).reshape(
test_size, num_verts, 3)
test_data_decoded_eigen = [p2e(m) for m in test_data_decoded]
### End of PCA version
else:
### Autoencoder
import keras
from keras.layers import Input, Dense
from keras.models import Model, load_model
import datetime
start_time = time.time()
train_size = num_samples
test_size = num_samples
test_data = numpy_displacements_sample[:test_size].reshape(
test_size, 3 * num_verts)
test_data_eigen = verts_sample[:test_size]
# numpy.random.shuffle(numpy_displacements_sample)
# train_data = numpy_verts_sample[test_size:test_size+train_size]
train_data = numpy_displacements_sample[0:train_size].reshape(
(train_size, 3 * num_verts))
mean = numpy.mean(train_data, axis=0)
std = numpy.std(train_data, axis=0)
mean = numpy.mean(train_data)
std = numpy.std(train_data)
s_min = numpy.min(train_data)
s_max = numpy.max(train_data)
def normalize(data):
return numpy.nan_to_num((data - mean) / std)
# return numpy.nan_to_num((train_data - s_min) / (s_max - s_min))
def denormalize(data):
return data * std + mean
# return data * (s_max - s_min) + s_min
train_data = normalize(train_data)
test_data = normalize(test_data)
# print(train_data)
# print(mean)
# print(std)
# exit()
# this is the size of our encoded representations
encoded_dim = 3
# Single autoencoder
# initializer = keras.initializers.RandomUniform(minval=0.0, maxval=0.01, seed=5)
# bias_initializer = initializer
activation = keras.layers.advanced_activations.LeakyReLU(
alpha=0.3) #'relu'
input = Input(shape=(len(train_data[0]), ))
output = input
output = Dense(30, activation=activation)(input)
output = Dense(512, activation=activation)(output)
output = Dense(64, activation=activation)(output)
output = Dense(encoded_dim, activation=activation,
name="encoded")(output)
output = Dense(64, activation=activation)(output)
output = Dense(512, activation=activation)(output)
output = Dense(30, activation=activation)(output)
output = Dense(len(train_data[0]), activation='linear')(
output
) #'linear',)(output) # First test seems to indicate no change on output with linear
autoencoder = Model(input, output)
optimizer = keras.optimizers.Adam(lr=0.001,
beta_1=0.9,
beta_2=0.999,
epsilon=1e-08,
decay=0)
autoencoder.compile(optimizer=optimizer, loss='mean_squared_error')
model_start_time = time.time()
autoencoder.fit(train_data,
train_data,
epochs=1000,
batch_size=num_samples,
shuffle=True,
validation_data=(test_data, test_data))
# output_path = 'trained_models/' + datetime.datetime.now().strftime("%I %M%p %B %d %Y") + '.h5'
# autoencoder.save(output_path)
print("Total model time: ", time.time() - model_start_time)
# Display
decoded_samples = denormalize(autoencoder.predict(test_data))
#decoded_samples = autoencoder.predict(test_data) * std + mean
test_data_decoded = (numpy_base_verts + decoded_samples).reshape(
test_size, num_verts, 3)
test_data_decoded_eigen = [p2e(m) for m in test_data_decoded]
### End of Autoencoder
# Error colours
error = numpy.sum((test_data_decoded - numpy_verts_sample)**2, axis=2)
colours = [igl.eigen.MatrixXd() for _ in range(num_samples)]
for i in range(num_samples):
igl.jet(p2e(error[i]), True, colours[i])
# Set up
viewer = igl.viewer.Viewer()
viewer.data.set_mesh(initial_verts, initial_faces)
def pre_draw(viewer):
global current_frame, verts_sample, show_decoded
if viewer.core.is_animating:
print(current_frame)
print(show_decoded)
if show_decoded:
viewer.data.set_vertices(
test_data_decoded_eigen[current_frame])
viewer.data.set_colors(colours[current_frame])
else:
viewer.data.set_vertices(test_data_eigen[current_frame])
viewer.data.compute_normals()
current_frame = (current_frame + 1) % test_size
return False
viewer.callback_pre_draw = pre_draw
viewer.callback_key_down = key_down
viewer.core.is_animating = False
# viewer.core.camera_zoom = 2.5
viewer.core.animation_max_fps = 30.0
viewer.launch()