- For #1999
- For GeoTrellis 1.0.0
There is a comparable performance in [1/3 - 1/1000000] of all points on aspect-tif.tif (256 x 256 | 512 x 512)
benchmarks in all three cases. It looks like the new CostDistance
implementation is more dependent on the exact points values. The more points the target tile has, the more inconsistent results we get, and the gap between the old implementation and the new one increases. 1/3 of points
and all points
(?) are the worst case