Pivot calibration: erroneous mixing algorithms by recursion

[Quentin Quadrat]

Hello, I'm not a good Python programmer, but it seems that the pivot_calibration_with_ransac function is calling badly its recursion:

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 11 in 549a534

 def pivot_calibration(tracking_matrices, configuration=None): 

will call

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 66 in 549a534

 return pivot_calibration_with_ransac( 

that will call

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 116 in 549a534

 def pivot_calibration_with_ransac(tracking_matrices, 

that will call

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 157 in 549a534

 pointer_offset, pivot_location, _ = pivot_calibration(sample) 

with implicit configuration=None as parameter that will call

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 55 in 549a534

 use_algebraic_one_step = True 

and

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 58 in 549a534

 return pivot_calibration_aos(tracking_matrices) 

but since

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 42 in 549a534

 use_ransac = False 

the

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 60 in 549a534

 if use_ransac: 

is not called

I've not traced calls (as noob in Python is this kind of tool exists ?), this is a simple "manual" code review, I may be wrong. I guess you have forgotten to pass configuration in the recursive calls. Finally, your recursive algorithm is complex to understand: I think pivot_calibration_with_ransac shall only call pivot_calibration_with_ransac but not the base function pivot_calibration.

[Stephen Thompson]

Thanks for raising this. I'm away this week so have assigned @mxochicale to take a look first. @mxochicale can you please take a look? @MattClarkson might be able to help too as I think he was the main author on this.

[Stephen Thompson]

I've had a look at this now. I think it is confusion caused by our overly complex recursion (which I don't think is actually recursion). My understanding is that pivot_calibration is a wrapper that will call one of three methods (pivot_calibration_aos, pivot_calibration_ransac, or pivot_calibration_sphere_fitting).
pivot_calibration_ransac calls pivot_calibration_aos multiple times to iteratively find an optimal solution (i.e. pivot_calibration_ransac is not meant to call itself (recurse)).
I've clarified that in the code see

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 157 in a1f4556

 pointer_offset, pivot_location, _ = pivot_calibration_aos(sample) 

and

scikit-surgerycalibration/sksurgerycalibration/algorithms/pivot.py

Line 183 in a1f4556

 pivot_calibration_aos(inlier_matrices) 

I hope that makes sense. If so I'll create a pull request for it.

[Quentin Quadrat]

My understanding is that pivot_calibration is a wrapper that will call one of three methods (pivot_calibration_aos, pivot_calibration_ransac, or pivot_calibration_sphere_fitting).

I agree with you.

pivot_calibration_ransac calls pivot_calibration_aos multiple times to iteratively find an optimal solution (i.e. pivot_calibration_ransac is not meant to call itself (recurse)).

I agree with you.

I hope that makes sense. If so I'll create a pull request for it.

The fix looks good to me. Probably you will have to also update unit tests.