Table 2.

Average ARI results for HDBSCAN(constraint) in comparison with other methods. HDBSCAN($\widehat{ϵ}$+constraint) refers to a combined version with HDBSCAN($\widehat{ϵ}$). Both versions were tested with (Alt.) and without (No Alt.) the option of choosing a better alternative from the hierarchy level below. HDB. (eom) refers to default unsupervised HDBSCAN. “Theoretically achievable” refers to the best HDBSCAN cluster selection possible w.r.t. the given hierarchy. All methods are based on Euclidean distance in x-y-v dimension.

Scene	Theoretically Achievable	HDBSCAN(Constraint)		HDBSCAN($\widehat{\mathit{ϵ}}$+Constraint)		HDB. (eom)	DBSCAN*
		No Alt.	Alt.	No Alt.	Alt.
1003	0.97	0.97	0.97	0.97	0.97	0.71	0.74
0239	0.90	0.89	0.90	0.89	0.90	0.48	0.74
0400	0.92	0.82	0.91	0.82	0.91	0.87	0.90
0553	0.96	0.83	0.86	0.88	0.91	0.71	0.83
Average	0.94	0.88	0.91	0.89	0.92	0.69	0.80