How to get point in moving image?

[sunxirui310]

Im trying to get a point warped in moving image. something goes wrong. someone give me a hand? thx!

class SimilarityTransformation(RigidTransformation):
    ......    
    def _get_transformation_matrix(self):
        self._compute_transformation()
        return self._compute_transformation_matrix()

in file 'affine_registration_2d.py' i added:

matT = transformation._get_transformation_matrix().data.cpu().numpy()
matT = np.vstack((matT,[0,0,1]))# make matT 3*3
matT_inv = np.linalg.inv(matT)# inv(matT)
...
#draw point in moving image
px=128
py=64
plt.subplot(131)
plt.imshow(fixed_image.numpy(), cmap='gray')
plt.plot(px, py, 'ro')
plt.title('Fixed Image')

plt.subplot(132)
plt.imshow(moving_image.numpy(), cmap='gray')
pxM=(px/256*matT_inv[0][0]+py/256*matT_inv[0][1]+matT_inv[0][2])*256
pyM=(px/256*matT_inv[1][0]+py/256*matT_inv[1][1]+matT_inv[1][2])*256
plt.plot(pxM,pyM, 'r+')
plt.title('Moving Image')

[RobinSandkuehler]

Hi,
sorry for the late answer. You can use

points_transformed = al.Points.transform(points, displacement)

to transform points. Hope this could help you.

[sunxirui310]

thx，i've got pytorch transform matrix -> opencv transform matrix . seems to make a detour.
just 2D works.

def theta2param(theta, w,h):
    if theta.shape == (2,3):
        theta = np.vstack((theta,[0,0,1]))
    theta = np.linalg.inv(theta)
    param = np.zeros([2,3])
    param[0,0] = theta[0,0]
    param[0,1] = theta[0,1] / h * w
    param[0,2] = (theta[0,2] - theta[0,0] - theta[0,1] + 1) * w / 2
    param[1,0] = theta[1,0] / w * h
    param[1,1] = theta[1,1]
    param[1,2] = (theta[1,2] - theta[1,0] - theta[1,1] + 1) * h / 2   
    return param

[RossoneriMaldini]

Hi,
sorry for the late answer. You can use

points_transformed = al.Points.transform(points, displacement)

to transform points. Hope this could help you.

Hi,
Thanks for sharing!

I'm trying to find corresponding point in registration. Could you please tell me the displacement meaning in your implementation? I check the value of displacement and find values between [-0.01,0.01] in both x and y direction(in kernel_registration_2d.py). So displacement not mean pixel translation in each direction?

I'm trying this way but encounter error...
Traceback (most recent call last):
File "kernel_registration_2d.py", line 139, in
main()
File "kernel_registration_2d.py", line 108, in main
points_transformed = al.Points.transform(pts1, displacement)
File "/home/share/kaka/ML/airlab/airlab/utils/points.py", line 113, in transform
transformed_points[i, :] = df_transform.TransformPoint(points[i, :])
File "/home/kaka/anaconda3/envs/pytorch/lib/python3.6/site-packages/SimpleITK/SimpleITK.py", line 5016, in TransformPoint
return _SimpleITK.Transform_TransformPoint(self, point)
TypeError: in method 'Transform_TransformPoint', argument 2 of type 'std::vector< double,std::allocator< double > > const &'

[yangxue0827]

my code:

displacement = transformation.get_displacement()
warped_image = al.transformation.utils.warp_image(moving_image, displacement)
polybox = [[654, 260], [654, 290], [700, 290], [700, 260]]
warped_polybox = al.Points.transform(np.array(polybox).reshape([-1, 2]), displacement)
displacement = al.create_displacement_image_from_image(displacement, moving_image)

error:

Traceback (most recent call last):
  File "/Users/yangxue/Desktop/yangxue/code/airlab/examples/demo.py", line 151, in <module>
    main()
  File "/Users/yangxue/Desktop/yangxue/code/airlab/examples/demo.py", line 99, in main
    warped_polybox = al.Points.transform(np.array(polybox).reshape([-1, 2]), displacement)
  File "/Users/yangxue/Desktop/yangxue/code/airlab/airlab/utils/points.py", line 105, in transform
    raise Exception("Datatype of displacement field not supported.")
Exception: Datatype of displacement field not supported.

So I used the wrong displacement. How do I get the correct displacement? Can you give an example?