faster laplace solver through image pyramids

[jordandekraker]

@akhanf has some code to do this already, check out his old nomatlab (I think?) branch

This is really only an issue when running with no initialization (ie. skip_template_shape_injection)

[akhanf]

I located where the snakefile with multi-res laplace is, it is actually in master already here:

https://github.com/khanlab/hippunfold/blob/master/hippunfold/resources/tpl-upenn/Snakefile

There is a rule to downsample coords:

hippunfold/hippunfold/resources/tpl-upenn/Snakefile

Line 26 in 49a2245

rule downsample_coords:

Run laplace without initialization at the lowest resolution:

hippunfold/hippunfold/resources/tpl-upenn/Snakefile

Line 37 in 49a2245

rule laplace_coords:

Then upsample coords:

hippunfold/hippunfold/resources/tpl-upenn/Snakefile

Line 72 in 49a2245

rule upsample_coords:

and run Laplace with init coords:

hippunfold/hippunfold/resources/tpl-upenn/Snakefile

Line 54 in 49a2245

rule laplace_init_lores:

The number of levels is defined by setting the wildcard constraints in the level wildcard, e.g. right now lowest res is level 2 (downsampled by 50% twice; level 1 is downsampled once, level 0 is original resolution)..

[akhanf]

speed-up by fast marching initialization #88 obviates the need for this right now, so closing this issue