def test2():
print("**** Testing 50% outliers with 500 points:")
N = 500
min_val = -2000
max_val = 3000
np.random.seed(7)
pts1 = np.random.random_sample(
(N, 2)).astype(np.float32) * (max_val - min_val) + min_val
# randomize theta, t_x, t_y
theta = np.float32(np.random.random_sample() * (0.1 - (-0.1)) + (-0.1))
t_x, t_y = np.random.random_sample(2).astype(
np.float32) * (max_val - min_val) + min_val
print("Actual transformation: Theta {}, Tx {}, Ty {}".format(
theta, t_x, t_y))
# apply the transformation
trans_mat = np.array([
[np.cos(theta), -np.sin(theta)],
[np.sin(theta), np.cos(theta)],
],
dtype=np.float32)
pts2 = np.dot(trans_mat, pts1.T).T + np.array([t_x, t_y])
# Add some noise
outlier_mask = np.random.choice(N, int(N / 2), replace=False)
pts2[outlier_mask] += np.random.normal(scale=30,
size=(len(outlier_mask), 2)) + 10
iterations = 100
epsilon = 5.0
min_inlier_ratio = 0.05
min_num_inlier = 0.1 * N
max_rot_deg = 0.1
st_time = time.time()
# out = ransac_cython.ransac(
# [pts1, pts2], [pts1, pts2],
# iterations,
# epsilon,
# min_inlier_ratio,
# min_num_inlier,
# 0, # det_delta
# 0, # max_stretch
# max_rot_deg,
# 0, # tri_angles_comparator
# )
out = ransac_cython.ransac_rigid(
#np.array([pts1.T, pts2.T]), np.array([pts1.T, pts2.T]),
[pts1, pts2],
[pts1, pts2],
iterations,
epsilon,
min_inlier_ratio,
min_num_inlier,
max_rot_deg)
end_time = time.time()
print("Output is: {}, time: {} seconds".format(out[1], end_time - st_time))
def test6():
print(
"**** Testing normal distribution with 50000 points, 50000 iterations:"
)
N = 50000
min_val = -2000
max_val = 3000
np.random.seed(7)
pts1 = np.random.random_sample(
(N, 2)).astype(np.float32) * (max_val - min_val) + min_val
# randomize theta, t_x, t_y
theta = np.float32(np.random.random_sample() * (0.1 - (-0.1)) + (-0.1))
t_x, t_y = np.random.random_sample(2).astype(
np.float32) * (max_val - min_val) + min_val
print("Actual transformation: Theta {}, Tx {}, Ty {}".format(
theta, t_x, t_y))
# apply the transformation
trans_mat = np.array([
[np.cos(theta), -np.sin(theta)],
[np.sin(theta), np.cos(theta)],
],
dtype=np.float32)
pts2 = np.dot(trans_mat, pts1.T).T + np.array([t_x, t_y])
# Add some noise
pts2 += np.random.normal(scale=3, size=(N, 2))
iterations = 50000
epsilon = 5.0
min_inlier_ratio = 0.05
min_num_inlier = 0.1 * N
max_rot_deg = 0.1
out = None
st_time = time.time()
# func = ransac_cython.ransac_rigid
# profile = line_profiler.LineProfiler(func)
# profile.runcall(func, np.array([pts1.T, pts2.T]), np.array([pts1.T, pts2.T]),
# iterations,
# epsilon,
# min_inlier_ratio,
# min_num_inlier,
# max_rot_deg)
# cProfile.runctx("ransac_cython.ransac_rigid(\
# np.array([pts1.T, pts2.T]), np.array([pts1.T, pts2.T]),\
# iterations,\
# epsilon,\
# min_inlier_ratio,\
# min_num_inlier,\
# max_rot_deg\
# )", globals(), locals(), "Profile.prof")
out = ransac_cython.ransac_rigid(
#np.array([pts1.T, pts2.T]), np.array([pts1.T, pts2.T]),
[pts1, pts2],
[pts1, pts2],
iterations,
epsilon,
min_inlier_ratio,
min_num_inlier,
max_rot_deg)
end_time = time.time()
print("Output is: {}, time: {} seconds".format(out[1], end_time - st_time))
# assert_stats(profile, func.__name__)
# s = pstats.Stats("Profile.prof")
# s.strip_dirs().sort_stats("time").print_stats()
print("Running original ransac - Very slow, uncomment if needed")