def iter_l2_attack_1_sampling_all(model, x, y, params):
epsilon = params["epsilon"]
n = params["n"]
k = params["k"]
kappa = params["kappa"]
tri_all_points = params["tri_all_points"]
epsilon = epsilon / float(n)
x_perturb = x
for i in range(n):
grad = model.grad_fn(x_perturb, y)
perturb = epsilon * grad / np.sqrt(np.sum(grad**2))
x_perturb = x_perturb + perturb
sort_idx = np.argsort(np.linalg.norm(x_perturb - x, axis=1))
perturbed = x_perturb[sort_idx[k:]]
border_points, border_triangles = alpha_shape_border(
x_perturb if tri_all_points else perturbed)
triangles = []
for tri in border_triangles:
triangles.append(border_points[tri])
sampled = farthest_point_sampling(np.array(triangles), None,
len(x_perturb), kappa)
return sampled
def example_alpha_shape(label="chair"):
np.random.seed(1234)
view_label = label
offset_idx = 0
num_points = 1024
f = h5py.File("../data/point_clouds.hdf5", "r")
shape_names = [line.rstrip() for line in open("../data/shape_names.txt")]
pc = f["points"][:][:, :num_points]
labels = f["labels"][:]
f.close()
print("Shape:", pc.shape)
print("Number of points:", num_points)
print("Labels:", [shape_names[idx] for idx in np.unique(labels)])
print("Selected label:", view_label)
match_idx = np.where(labels == shape_names.index(view_label))[0]
view_pc = pc[match_idx[offset_idx]]
print("Shape index:", match_idx[offset_idx])
plt.figure(figsize = (30, 15))
plt.subplot(121, projection = "3d")
plt.gca().scatter(*view_pc.T, zdir = "y", s = 20, c = view_pc.T[1], cmap = "winter")
scale_plot()
plt.subplot(122, projection = "3d")
alpha_points, alpha_triangles = alpha_shape_border(view_pc, alpha_std = 0.0, epsilon = 0.001)
alpha_points = alpha_points[:, (0, 2, 1)]
print("Number of points in alpha shape:", alpha_points.shape[0])
print("Number of triangles in alpha shape:", alpha_triangles.shape[0])
plt.gca().plot_trisurf(*alpha_points.T, triangles = alpha_triangles, cmap = "winter")
scale_plot()
plt.subplots_adjust(left = 0, bottom = 0, right = 1, top = 1, wspace = 0, hspace = 0)
plt.show()
def __init__(self, x, alpha_std = 0.0, thickness = 0.0):
self.thickness = thickness
# construct the bounding triangles of the points
border_points, border_tri = alpha_shape_border(x, alpha_std = alpha_std)
triangles, tri_center, self.max_radius = _calc_tri_center(border_points, border_tri)
triangles = np.array(triangles)
tri_center = np.vstack(tri_center)
self.center = np.empty((len(triangles) * 2, 3))
self.radius_lo = np.empty(len(triangles) * 2)
self.radius_hi = np.empty(len(triangles) * 2)
self.inside_node = np.empty(len(triangles) * 2, dtype = int)
self.outside_node = np.empty(len(triangles) * 2, dtype = int)
self.triangle = np.empty((len(triangles) * 2, 3, 3))
self.is_leaf = np.empty(len(triangles) * 2, dtype = bool)
self.curr_idx = 0
self.root = self.build(triangles, tri_center)
def __init__(self, x, alpha_std = 0.0, thickness = 0.0):
self.thickness = thickness
# construct the bounding triangles of the points
border_points, border_tri = alpha_shape_border(x, alpha_std = alpha_std)
triangles = []
tri_center = []
self.max_radius = 0.0
for tri in border_tri:
# get the minimum bounding sphere of each triangle
tri = border_points[tri]
center, radius = bounding_sphere(tri)
triangles.append(tri)
tri_center.append(center)
self.max_radius = max(self.max_radius, radius)
triangles = np.array(triangles)
tri_center = np.vstack(tri_center)
self.root = self.build(triangles, tri_center)
def iter_l2_attack_n_sampling_rbf(model, x, y, params):
epsilon = params["epsilon"]
n = params["n"]
k = params["k"]
kappa = params["kappa"]
num_farthest = params["num_farthest"]
shape = params["shape"]
epsilon = epsilon / float(n)
x_perturb = x
for i in range(n):
grad = model.grad_fn(x_perturb, y)
perturb = epsilon * grad / np.sqrt(np.sum(grad**2))
x_perturb = x_perturb + perturb
sort_idx = np.argsort(np.linalg.norm(x_perturb - x, axis=1))
perturbed = x_perturb[sort_idx[k:]]
border_points, border_triangles = alpha_shape_border(x_perturb)
triangles = []
for tri in border_triangles:
triangles.append(border_points[tri])
sampled = radial_basis_sampling(np.array(triangles), perturbed, k,
kappa, num_farthest, shape)
x_sample = np.empty((len(x_perturb), 3))
for i in range(len(sort_idx)):
if i < k:
x_sample[sort_idx[i]] = sampled[i]
else:
x_sample[sort_idx[i]] = x_perturb[sort_idx[i]]
x_perturb = x_sample
return x_perturb
def iter_l2_attack_n_sampling(model, x, y, params):
epsilon = float(params["epsilon"])
n = int(params["n"])
k = int(params["k"])
kappa = int(params["kappa"])
tri_all_points = str(params["tri_all_points"]) == "True"
epsilon = epsilon / float(n)
x_perturb = x
for i in range(n):
grad = model.grad_fn(x_perturb, y)
perturb = epsilon * grad / np.sqrt(np.sum(grad**2))
x_perturb = x_perturb + perturb
sort_idx = np.argsort(np.linalg.norm(x_perturb - x, axis=1))
perturbed = x_perturb[sort_idx[k:]]
border_points, border_triangles = alpha_shape_border(
x_perturb if tri_all_points else perturbed)
triangles = []
for tri in border_triangles:
triangles.append(border_points[tri])
sampled = farthest_point_sampling(np.array(triangles), perturbed, k,
kappa)
x_sample = np.empty((len(x_perturb), 3))
for i in range(len(sort_idx)):
if i < k:
x_sample[sort_idx[i]] = sampled[i]
else:
x_sample[sort_idx[i]] = x_perturb[sort_idx[i]]
x_perturb = x_sample
return x_perturb
view_pc = pc[match_idx[offset_idx]]
print("Shape index:", match_idx[offset_idx])
plt.figure(figsize=(30, 15))
def scale_plot():
plt.gca().auto_scale_xyz((-1, 1), (-1, 1), (-1, 1))
plt.gca().view_init(0, 0)
plt.axis("off")
plt.subplot(121, projection="3d")
alpha_points, alpha_triangles = alpha_shape_border(view_pc)
alpha_points = alpha_points[:, (0, 2, 1)]
print("Number of points in alpha shape:", alpha_points.shape[0])
print("Number of triangles in alpha shape:", alpha_triangles.shape[0])
plt.gca().plot_trisurf(*alpha_points.T, triangles=alpha_triangles)
rand_perturb = (np.random.random(view_pc.shape) * 0.03 +
0.03) * np.random.choice(np.array([-1, 1]), size=view_pc.shape)
perturbed = view_pc + rand_perturb
colors = np.random.random(num_points)
plt.gca().scatter(*perturbed.T, zdir="y", s=300, c=colors, cmap="rainbow")
scale_plot()
print("Shape index:", match_idx[offset_idx])
plt.figure(figsize = (30, 15))
def scale_plot():
plt.gca().auto_scale_xyz((-1, 1), (-1, 1), (-1, 1))
plt.gca().view_init(0, 0)
plt.axis("off")
plt.subplot(121, projection = "3d")
plt.gca().scatter(*view_pc.T, zdir = "y", s = 20, c = view_pc.T[1], cmap = "winter")
scale_plot()
plt.subplot(122, projection = "3d")
alpha_points, alpha_triangles = alpha_shape_border(view_pc, alpha_std = 0.0, epsilon = 0.001)
alpha_points = alpha_points[:, (0, 2, 1)]
print("Number of points in alpha shape:", alpha_points.shape[0])
print("Number of triangles in alpha shape:", alpha_triangles.shape[0])
plt.gca().plot_trisurf(*alpha_points.T, triangles = alpha_triangles, cmap = "winter")
scale_plot()
plt.subplots_adjust(left = 0, bottom = 0, right = 1, top = 1, wspace = 0, hspace = 0)
plt.show()
# plt.savefig("", bbox_inches = 0)
def example_perturb_proj_tree(label="chair"):
np.random.seed(1234)
view_label = label
offset_idx = 0
num_points = 1024
f = h5py.File("../data/point_clouds.hdf5", "r")
shape_names = [line.rstrip() for line in open("../data/shape_names.txt")]
pc = f["points"][:][:, :num_points]
labels = f["labels"][:]
f.close()
print("Shape:", pc.shape)
print("Number of points:", num_points)
print("Labels:", [shape_names[idx] for idx in np.unique(labels)])
print("Selected label:", view_label)
match_idx = np.where(labels == shape_names.index(view_label))[0]
view_pc = pc[match_idx[offset_idx]]
print("Shape index:", match_idx[offset_idx])
plt.figure(figsize=(30, 15))
plt.subplot(121, projection="3d")
alpha_points, alpha_triangles = alpha_shape_border(view_pc)
alpha_points = alpha_points[:, (0, 2, 1)]
print("Number of points in alpha shape:", alpha_points.shape[0])
print("Number of triangles in alpha shape:", alpha_triangles.shape[0])
plt.gca().plot_trisurf(*alpha_points.T, triangles=alpha_triangles)
rand_perturb = (np.random.random(view_pc.shape) * 0.03 +
0.03) * np.random.choice(np.array([-1, 1]),
size=view_pc.shape)
perturbed = view_pc + rand_perturb
colors = np.random.random(num_points)
plt.gca().scatter(*perturbed.T, zdir="y", s=300, c=colors, cmap="rainbow")
scale_plot()
plt.subplot(122, projection="3d")
plt.gca().plot_trisurf(*alpha_points.T, triangles=alpha_triangles)
print("Projection started.")
start_time = time.time()
tree = PerturbProjTree(view_pc, thickness=0.01)
projected = tree.project(perturbed, rand_perturb)
end_time = time.time()
print("Projection ended in %f seconds." % (end_time - start_time))
plt.gca().scatter(*projected.T, zdir="y", s=300, c=colors, cmap="rainbow")
scale_plot()
plt.subplots_adjust(left=0, bottom=0, right=1, top=1, wspace=0, hspace=0)
plt.show()