def test_smoke_multi_batch(self, device, dtype):
points_2d = torch.zeros(2, 3, 2, device=device, dtype=dtype)
depth = torch.ones(2, 3, 1, device=device, dtype=dtype)
camera_matrix = torch.eye(3, device=device,
dtype=dtype).expand(2, 3, -1, -1)
point_3d = kornia.unproject_points(points_2d, depth, camera_matrix)
assert point_3d.shape == (2, 3, 3)
def test_unproject_center(self):
point_2d = torch.tensor([[0., 0.]])
depth = torch.tensor([[2.]])
camera_matrix = torch.tensor([
[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.],
])
expected = torch.tensor([[0., 0., 2.]])
actual = kornia.unproject_points(point_2d, depth, camera_matrix)
assert_allclose(actual, expected)
def test_project_and_unproject(self):
point_3d = torch.tensor([[10., 2., 30.]])
depth = point_3d[..., -1:]
camera_matrix = torch.tensor([[
[2746., 0., 991.],
[0., 2748., 619.],
[0., 0., 1.],
]])
point_2d = kornia.project_points(point_3d, camera_matrix)
point_3d_hat = kornia.unproject_points(point_2d, depth, camera_matrix)
assert_allclose(point_3d, point_3d_hat)
def test_unproject_and_project(self):
point_2d = torch.tensor([[0., 0.]])
depth = torch.tensor([[2.]])
camera_matrix = torch.tensor([
[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.],
])
point_3d = kornia.unproject_points(point_2d, depth, camera_matrix)
point_2d_hat = kornia.project_points(point_3d, camera_matrix)
assert_allclose(point_2d, point_2d_hat)
def test_unproject_center_normalize(self, device):
point_2d = torch.tensor([[0., 0.]]).to(device)
depth = torch.tensor([[2.]]).to(device)
camera_matrix = torch.tensor([
[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.],
]).to(device)
expected = torch.tensor([[0., 0., 2.]]).to(device)
actual = kornia.unproject_points(point_2d, depth, camera_matrix, True)
assert_allclose(actual, expected)
def test_jit(self):
@torch.jit.script
def op_script(points_2d, depth, camera_matrix):
return kornia.unproject_points(points_2d, depth, camera_matrix,
False)
points_2d = torch.zeros(1, 2)
depth = torch.ones(1, 1)
camera_matrix = torch.eye(3).expand(1, -1, -1)
actual = op_script(points_2d, depth, camera_matrix)
expected = kornia.unproject_points(points_2d, depth, camera_matrix)
assert_allclose(actual, expected)
def test_unproject_center(self, device, dtype):
point_2d = torch.tensor([[0., 0.]], device=device, dtype=dtype)
depth = torch.tensor([[2.]], device=device, dtype=dtype)
camera_matrix = torch.tensor([
[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.],
],
device=device,
dtype=dtype)
expected = torch.tensor([[0., 0., 2.]], device=device, dtype=dtype)
actual = kornia.unproject_points(point_2d, depth, camera_matrix)
assert_allclose(actual, expected, atol=1e-4, rtol=1e-4)
def test_unproject_and_project(self, device, dtype):
point_2d = torch.tensor([[0., 0.]], device=device, dtype=dtype)
depth = torch.tensor([[2.]], device=device, dtype=dtype)
camera_matrix = torch.tensor([
[1., 0., 0.],
[0., 1., 0.],
[0., 0., 1.],
],
device=device,
dtype=dtype)
point_3d = kornia.unproject_points(point_2d, depth, camera_matrix)
point_2d_hat = kornia.project_points(point_3d, camera_matrix)
assert_allclose(point_2d, point_2d_hat, atol=1e-4, rtol=1e-4)
def test_jit(self, device, dtype):
@torch.jit.script
def op_script(points_2d, depth, camera_matrix):
return kornia.unproject_points(points_2d, depth, camera_matrix,
False)
points_2d = torch.zeros(1, 2, device=device, dtype=dtype)
depth = torch.ones(1, 1, device=device, dtype=dtype)
camera_matrix = torch.eye(3, device=device,
dtype=dtype).expand(1, -1, -1)
actual = op_script(points_2d, depth, camera_matrix)
expected = kornia.unproject_points(points_2d, depth, camera_matrix)
assert_allclose(actual, expected, atol=1e-4, rtol=1e-4)
def samplePosition(self, num_samples=1):
"""
Returns
-------
t : torch.Tensor of shape (num_samples, 3)
"""
camera_dist = self.renderer.t[0, 0, 2]
px_bounds = torch.tensor([[0, 0], [self.image_size, self.image_size]],
dtype=torch.float).cuda()
depth_coords = torch.full(px_bounds.shape[0:1],
camera_dist)[..., None].cuda()
xy_bounds = kornia.unproject_points(px_bounds, depth_coords,
self.renderer.K)[:, :-1]
xy_samples = _sample_uniform(num_samples, xy_bounds[0], xy_bounds[1])
z_samples = torch.full((num_samples, 1), 0).cuda()
t = torch.cat((xy_samples, z_samples), dim=1)
return t.cuda()
def test_smoke_batch(self):
points_2d = torch.zeros(2, 2)
depth = torch.ones(2, 1)
camera_matrix = torch.eye(3).expand(2, -1, -1)
point_3d = kornia.unproject_points(points_2d, depth, camera_matrix)
assert point_3d.shape == (2, 3)
def op_script(points_2d, depth, camera_matrix):
return kornia.unproject_points(points_2d, depth, camera_matrix,
False)
def test_smoke(self, device):
points_2d = torch.zeros(1, 2).to(device)
depth = torch.ones(1, 1).to(device)
camera_matrix = torch.eye(3).expand(1, -1, -1).to(device)
point_3d = kornia.unproject_points(points_2d, depth, camera_matrix)
assert point_3d.shape == (1, 3)
