gfleishman / circuitseeker Goto Github PK

Hi Greg,

I have been trying to optimize the parameters space to get single-cell resolution in an EM/2P pair in a fully automated way. I have attached an image that shows the output.

Here you can see that some swirls appear (left red circle) although the shearing values have been kepts to 0. This tends to appear at high and low volume dicing.
Also, conversely to the previous versions erronous deformations tend to appear when overlapping data is missing rather that keeping the original deformation (left red circle). The control over this parameter seem to be gone in the last posted version.

I have used the following parameters:
block_schedule = [ [tuple(np.round(np.array(fix.shape) / 64).astype(int))], [tuple(np.round(np.array(fix.shape) / 32).astype(int))]]
randomiter=2500

parameter_schedule=
[{'random_kwargs':{'max_translation':max_trans, 'max_rotation':2. * np.pi/180., 'max_scale':1.02, 'max_shear':.00, 'random_iterations':randomiter, 'affine_align_best':10, 'iterations':24,}, 'affine_kwargs':{}, 'deform_kwargs':{'control_point_spacing':5.0, 'control_point_levels':[2, 4, 6], 'iterations':100, 'metric':'MI',}, },
{'random_kwargs':{'max_translation':max_trans, 'max_rotation':2. * np.pi/180., 'max_scale':1.02, 'max_shear':.00, 'random_iterations':randomiter, 'affine_align_best':10, 'iterations':24,}, 'affine_kwargs':{}, 'deform_kwargs':{'control_point_spacing':5.0, 'control_point_levels':[2, 4], 'iterations':100, 'metric':'MI',}, } ]

I am sorry for this overwhelming message ...
Happy to give you more clarifications to reproduce this issue.

Thanks a lot,
Jon

Hello Greg,

Could you please provide a notebook example to use the new alignment_pipeline function
This new input has changed is a bit tricky:
steps : list of tuples in this form [(str, dict), (str, dict), ...]

Thanks a lot,
Jon