Talus's evaluation method divides all distance metrics by the distance between two disjoint sets of real maps as a noise baseline, where 1.0 means the generated results are indistinguishable from real maps.