def test_check_corner_case(self, device, dtype):
trans = torch.tensor(
[[[0.0, 0.0, 1.0], [0.0, 2.0, 0.0], [0.5, 0.0, 0.0]]],
device=device,
dtype=dtype)
trans_expected = trans.clone()
trans_norm = epi.normalize_transformation(trans)
assert_close(trans_norm, trans_expected, atol=1e-4, rtol=1e-4)
def test_check_last_val(self, device, dtype):
trans = torch.tensor(
[[[0.0, 0.0, 1.0], [0.0, 2.0, 0.0], [0.5, 0.0, 0.5]]],
device=device,
dtype=dtype)
trans_expected = torch.tensor(
[[[0.0, 0.0, 2.0], [0.0, 4.0, 0.0], [1.0, 0.0, 1.0]]],
device=device,
dtype=dtype)
trans_norm = epi.normalize_transformation(trans)
assert_allclose(trans_norm, trans_expected, atol=1e-4, rtol=1e-4)
def generate_two_view_random_scene(device: torch.device = torch.device("cpu"),
dtype: torch.dtype = torch.float32
) -> Dict[str, torch.Tensor]:
num_views: int = 2
num_points: int = 30
scene: Dict[str, torch.Tensor] = epi.generate_scene(num_views, num_points)
# internal parameters (same K)
K1 = scene['K'].to(device, dtype)
K2 = K1.clone()
# rotation
R1 = scene['R'][0:1].to(device, dtype)
R2 = scene['R'][1:2].to(device, dtype)
# translation
t1 = scene['t'][0:1].to(device, dtype)
t2 = scene['t'][1:2].to(device, dtype)
# projection matrix, P = K(R|t)
P1 = scene['P'][0:1].to(device, dtype)
P2 = scene['P'][1:2].to(device, dtype)
# fundamental matrix
F_mat = epi.fundamental_from_projections(P1[..., :3, :], P2[..., :3, :])
F_mat = epi.normalize_transformation(F_mat)
# points 3d
X = scene['points3d'].to(device, dtype)
# projected points
x1 = scene['points2d'][0:1].to(device, dtype)
x2 = scene['points2d'][1:2].to(device, dtype)
return dict(K1=K1,
K2=K2,
R1=R1,
R2=R2,
t1=t1,
t2=t2,
P1=P1,
P2=P2,
F=F_mat,
X=X,
x1=x1,
x2=x2)
def test_from_to_projections(self, device, dtype):
P1 = torch.tensor([[
[1., 0., 0., 0.],
[0., 1., 0., 0.],
[1., 0., 1., 0.],
]],
device=device,
dtype=dtype)
P2 = torch.tensor([[
[1., 1., 1., 3.],
[0., 2., 0., 3.],
[0., 1., 1., 0.],
]],
device=device,
dtype=dtype)
F_mat = epi.fundamental_from_projections(P1, P2)
P_mat = epi.projections_from_fundamental(F_mat)
F_hat = epi.fundamental_from_projections(P_mat[..., 0], P_mat[..., 1])
F_mat_norm = epi.normalize_transformation(F_mat)
F_hat_norm = epi.normalize_transformation(F_hat)
assert_allclose(F_mat_norm, F_hat_norm)
def test_shape(self, batch_size, rows, cols, device, dtype):
B, N, M = batch_size, rows, cols
trans = torch.rand(B, N, M, device=device, dtype=dtype)
trans_norm = epi.normalize_transformation(trans)
assert trans_norm.shape == (B, N, M)
def test_smoke(self, device, dtype):
trans = torch.rand(2, 2, device=device, dtype=dtype)
trans_norm = epi.normalize_transformation(trans)
assert trans_norm.shape == (2, 2)
