WIP: SyN landmarks.

[ntustison]

Hey @stnava , can you check this when you have some time? Below is a toy example I've been using. Note that the return value is a list of transforms with transforms[0] being the forward transform and transforms[1] being the reverse transform.

import ants
import numpy as np
domain_image = ants.image_read(ants.get_ants_data("r16"))
grid = ants.create_warped_grid(domain_image, grid_directions=(True, True))

fixed = np.array([[50.0,50.0],[200.0,50.0],[50.0,200.0]])
moving = np.array([[75.0,75.0],[175.0,75.0],[75.0,175.0]])
xfrms = ants.fit_transform_to_paired_points(moving, fixed, transform_type="syn", domain_image=domain_image, 
                 mesh_size=10, number_of_fitting_levels=6, number_of_compositions=100, composition_step_size=0.1)
warped_grid = xfrms[0].apply_to_image(grid)
ants.plot(warped_grid)
warped_grid = xfrms[1].apply_to_image(grid)
ants.plot(warped_grid)

[stnava]

looks good - here is the example I used as a grad student:

fixed = np.array([[50.0,50.0],[200.0,50.0],[200.0,200.0]])
moving = np.array([[50.0,50.0],[50.0,200.0],[200.0,200.0]])

[stnava]

my example "breaks" the inversion .... it used to work with the "uni-directional" option ie with what was called "diffeo" ... I think "diffeo" is now broken.

since you are working on this, I would like to request a

fwd, inv = ants.integrate_velocity_field( field, time_step  )

function ... that will let us do very related things but with direct optimization of the flow field .... I also think it will perform better for these types of cases ... although this example is an extreme case, it would be nice to have something that worked for it.

[ntustison]

Great idea, @stnava. I'll work on that.

So I'll merge this since it appears to be working. Also, I discovered the "brokenness" of the diffeo option earlier and it's included in this merge. Once you update, let me know if it's fixed or if I need to dig further.