def test_smoke(self, device, dtype):
points = torch.rand(1, 2, device=device, dtype=dtype)
K = torch.rand(3, 3, device=device, dtype=dtype)
distCoeff = torch.rand(4, device=device, dtype=dtype)
pointsu = undistort_points(points, K, distCoeff)
assert points.shape == pointsu.shape
new_K = torch.rand(3, 3, device=device, dtype=dtype)
pointsu = undistort_points(points, K, distCoeff, new_K)
assert points.shape == pointsu.shape
def test_shape(self, batch_size, num_points, num_distcoeff, device, dtype):
B, N, Ndist = batch_size, num_points, num_distcoeff
points = torch.rand(B, N, 2, device=device, dtype=dtype)
K = torch.rand(B, 3, 3, device=device, dtype=dtype)
distCoeff = torch.rand(B, Ndist, device=device, dtype=dtype)
pointsu = undistort_points(points, K, distCoeff)
assert pointsu.shape == (B, N, 2)
new_K = torch.rand(B, 3, 3, device=device, dtype=dtype)
pointsu = undistort_points(points, K, distCoeff, new_K)
assert pointsu.shape == (B, N, 2)
def test_opencv_five_coeff(self, device, dtype):
# Test using 5 distortion coefficients
pts = torch.tensor([[1028.0374, 788.7520], [1025.1218, 716.8726],
[1022.1792, 645.1857]],
device=device,
dtype=dtype)
K = torch.tensor(
[[1.7315e03, 0.0000e00, 6.2289e02],
[0.0000e00, 1.7320e03, 5.3537e02],
[0.0000e00, 0.0000e00, 1.0000e00]],
device=device,
dtype=dtype,
)
dist = torch.tensor([-0.1007, 0.2650, -0.0018, 0.0007, -0.2597],
device=device,
dtype=dtype)
# Expected ouput generated with OpenCV:
# import cv2
# ptsu_expected = cv2.undistortPoints(pts.numpy().reshape(-1,1,2), K.numpy(),
# dist1.numpy(), None, None, K.numpy()).reshape(-1,2)
ptsu_expected = torch.tensor(
[[1030.5992, 790.65533], [1027.3059, 718.10020],
[1024.0700, 645.90600]],
device=device,
dtype=dtype)
ptsu = undistort_points(pts, K, dist)
assert_allclose(ptsu, ptsu_expected, rtol=1e-4, atol=1e-4)
def test_opencv_all_coeff(self, device, dtype):
# Test using 14 distortion coefficients
pts = torch.tensor([[1028.0374, 788.7520], [1025.1218, 716.8726],
[1022.1792, 645.1857]],
device=device,
dtype=dtype)
K = torch.tensor(
[[1.7315e03, 0.0000e00, 6.2289e02],
[0.0000e00, 1.7320e03, 5.3537e02],
[0.0000e00, 0.0000e00, 1.0000e00]],
device=device,
dtype=dtype,
)
dist = torch.tensor(
[
-5.6388e-02,
2.3881e-01,
8.3374e-02,
2.0710e-03,
7.1349e00,
5.6335e-02,
-3.1738e-01,
4.9981e00,
-4.0287e-03,
-2.8246e-02,
-8.6064e-02,
1.5543e-02,
-1.7322e-01,
2.3154e-03,
],
device=device,
dtype=dtype,
)
# Expected ouput generated with OpenCV:
# import cv2
# ptsu_expected = cv2.undistortPoints(pts.numpy().reshape(-1,1,2), K.numpy(),
# dist2.numpy(), None, None, K.numpy()).reshape(-1,2)
ptsu_expected = torch.tensor(
[[1030.8245, 786.3807], [1027.5505, 715.0732],
[1024.2753, 644.0319]],
device=device,
dtype=dtype)
ptsu = undistort_points(pts, K, dist)
assert_allclose(ptsu, ptsu_expected, rtol=1e-4, atol=1e-4)
def test_opencv_stereo(self, device, dtype):
# Udistort stereo points with data given in two batches using 14 distortion coefficients
pts = torch.tensor(
[
[[1028.0374, 788.7520], [1025.1218, 716.8726],
[1022.1792, 645.1857]],
[[345.9135, 847.9113], [344.0880, 773.9890],
[342.2381, 700.3029]],
],
device=device,
dtype=dtype,
)
K = torch.tensor(
[
[
[3.3197e03, 0.0000e00, 6.1813e02],
[0.0000e00, 3.3309e03, 5.2281e02],
[0.0000e00, 0.0000e00, 1.0000e00],
],
[
[1.9206e03, 0.0000e00, 6.1395e02],
[0.0000e00, 1.9265e03, 7.7164e02],
[0.0000e00, 0.0000e00, 1.0000e00],
],
],
device=device,
dtype=dtype,
)
dist = torch.tensor(
[
[
-5.6388e-02,
2.3881e-01,
8.3374e-02,
2.0710e-03,
7.1349e00,
5.6335e-02,
-3.1738e-01,
4.9981e00,
-4.0287e-03,
-2.8246e-02,
-8.6064e-02,
1.5543e-02,
-1.7322e-01,
2.3154e-03,
],
[
1.4050e-03,
-3.0691e00,
-1.0209e-01,
-2.3687e-02,
-1.7082e02,
4.3593e-03,
-3.1904e00,
-1.7050e02,
1.7854e-02,
1.8999e-02,
9.9122e-02,
3.6675e-02,
3.0816e-03,
-5.7133e-02,
],
],
device=device,
dtype=dtype,
)
# Expected ouput generated with OpenCV:
# import cv2
# ptsu_expected1 = cv2.undistortPoints(pts[0].numpy().reshape(-1,1,2), K[0].numpy(),
# dist[0].numpy(), None, None, K[0].numpy()).reshape(-1,2)
# ptsu_expected2 = cv2.undistortPoints(pts[1].numpy().reshape(-1,1,2), K[1].numpy(),
# dist[1].numpy(), None, None, K[1].numpy()).reshape(-1,2)
ptsu_expected1 = torch.tensor(
[[1029.3234, 785.4813], [1026.1599, 714.3689],
[1023.02045, 643.5359]],
device=device,
dtype=dtype)
ptsu_expected2 = torch.tensor(
[[344.04456, 848.7696], [344.27606, 774.1254],
[344.47018, 700.8522]],
device=device,
dtype=dtype)
ptsu = undistort_points(pts, K, dist)
assert_allclose(ptsu[0], ptsu_expected1, rtol=1e-4, atol=1e-4)
assert_allclose(ptsu[1], ptsu_expected2, rtol=1e-4, atol=1e-4)
