Closed masadcv closed 1 year ago
Using the code from #40 with modification to add GeodisTK results, here is comparison of 3d Euclidean distance with the bug:
And here are results with fixes from this PR:
With the PR, the Euclidean distances improve and is in-line with Euclidean distance from previous raster scan methods (e.g. using GeodisTK). It, however, is an approximation and unavoidable artifacts are present which are due reliance of these methods on iterative raster scans for distance transform computation
As reported in #40 , the only Euclidean case appeared to be computing taxicab distance.
On closer look, it came down to bug in local distance calculation
This PR attempts to fix this.