def test_unbatched(self, num_points, device, dtype):
N = num_points
E_mat = torch.rand(3, 3, device=device, dtype=dtype)
K1 = torch.rand(3, 3, device=device, dtype=dtype)
K2 = torch.rand(3, 3, device=device, dtype=dtype)
x1 = torch.rand(N, 2, device=device, dtype=dtype)
x2 = torch.rand(N, 2, device=device, dtype=dtype)
R, t, X = epi.motion_from_essential_choose_solution(
E_mat, K1, K2, x1[1:-1, :], x2[1:-1, :])
assert R.shape == (3, 3)
assert t.shape == (3, 1)
assert X.shape == (N - 2, 3)
mask = torch.zeros(N, dtype=torch.bool, device=device)
mask[1:-1] = True
Rm, tm, Xm = epi.motion_from_essential_choose_solution(E_mat,
K1,
K2,
x1,
x2,
mask=mask)
assert_allclose(R, Rm)
assert_allclose(t, tm)
assert_allclose(X, Xm[1:-1, :])
def test_masking(self, device, dtype):
E_mat = torch.rand(2, 3, 3, device=device, dtype=dtype)
K1 = torch.rand(2, 3, 3, device=device, dtype=dtype)
K2 = torch.rand(2, 3, 3, device=device, dtype=dtype)
x1 = torch.rand(2, 10, 2, device=device, dtype=dtype)
x2 = torch.rand(2, 10, 2, device=device, dtype=dtype)
R, t, X = epi.motion_from_essential_choose_solution(E_mat, K1, K2, x1[:, 1:-1, :], x2[:, 1:-1, :])
mask = torch.zeros(2, 10, dtype=torch.bool, device=device)
mask[:, 1:-1] = True
Rm, tm, Xm = epi.motion_from_essential_choose_solution(E_mat, K1, K2, x1, x2, mask=mask)
assert_close(R, Rm)
assert_close(t, tm)
assert_close(X, Xm[:, 1:-1, :])
def test_smoke(self, device, dtype):
E_mat = torch.rand(1, 3, 3, device=device, dtype=dtype)
K1 = torch.rand(1, 3, 3, device=device, dtype=dtype)
K2 = torch.rand(1, 3, 3, device=device, dtype=dtype)
x1 = torch.rand(1, 1, 2, device=device, dtype=dtype)
x2 = torch.rand(1, 1, 2, device=device, dtype=dtype)
R, t, X = epi.motion_from_essential_choose_solution(E_mat, K1, K2, x1, x2)
assert R.shape == (1, 3, 3)
assert t.shape == (1, 3, 1)
assert X.shape == (1, 1, 3)
def test_shape(self, batch_size, num_points, device, dtype):
B, N = batch_size, num_points
E_mat = torch.rand(B, 3, 3, device=device, dtype=dtype)
K1 = torch.rand(B, 3, 3, device=device, dtype=dtype)
K2 = torch.rand(1, 3, 3, device=device, dtype=dtype) # check for broadcasting
x1 = torch.rand(B, N, 2, device=device, dtype=dtype)
x2 = torch.rand(B, 1, 2, device=device, dtype=dtype) # check for broadcasting
R, t, X = epi.motion_from_essential_choose_solution(E_mat, K1, K2, x1, x2)
assert R.shape == (B, 3, 3)
assert t.shape == (B, 3, 1)
assert X.shape == (B, N, 3)
def test_two_view(self, device, dtype):
scene = utils.generate_two_view_random_scene(device, dtype)
E_mat = epi.essential_from_Rt(scene['R1'], scene['t1'], scene['R2'], scene['t2'])
R, t = epi.relative_camera_motion(scene['R1'], scene['t1'], scene['R2'], scene['t2'])
t = torch.nn.functional.normalize(t, dim=1)
R_hat, t_hat, _ = epi.motion_from_essential_choose_solution(
E_mat, scene['K1'], scene['K2'], scene['x1'], scene['x2']
)
assert_close(t, t_hat)
assert_close(R, R_hat, rtol=1e-4, atol=1e-4)
