def farthest_point_idx(sampled_points, initial_points, num_points):
curr_points = np.empty(num_points, dtype=np.int64)
dists = np.full(len(sampled_points), np.inf)
if initial_points is None:
dists = np.minimum(
dists, norm(sampled_points - sampled_points[0].reshape((1, -1))))
curr_points[0] = 0
start_idx = 1
else:
for i in range(len(initial_points)):
dists = np.minimum(
dists,
norm(sampled_points - initial_points[i].reshape((1, -1))))
start_idx = 0
for i in range(start_idx, num_points):
curr_points[i] = np.argmax(dists)
dists = np.minimum(
dists,
norm(sampled_points -
sampled_points[curr_points[i]].reshape((1, -1))))
return curr_points
def sample_on_line_segments(x, x_perturb, sigma):
small_perturb = 0.003
norms = norm(x_perturb - x)
prefix = []
for i in range(len(norms)):
if i == 0:
prefix.append(norms[i])
else:
prefix.append(prefix[i - 1] + norms[i])
total_prob = prefix[-1]
count = np.zeros(len(norms))
for i in range(sigma):
rand = np.random.uniform(0.0, total_prob)
idx = binary_search(prefix, rand)
count[idx] += 1.0
x_sample = np.empty((sigma, 3))
idx = 0
for i in range(len(norms)):
for j in range(count[i]):
x_sample[idx] = x[i] + (
x_perturb[i] -
x[i]) * j / count[i] + small_perturb * np.random.randn(3)
idx += 1
return x_sample