Table 7.

Comparison of different variants of our method on Scene8. The best performance indicators of all the taxonomies are marked as bold.

	ori.	cri. 1	cri. 2	cri. 3	cri. 4	cri. 5	$\overline{\mathit{\xi }}$ ± s
with all components	1.64	1.73	1.67	1.75	1.86	2.05	1.78 ± 0.14
No log-ratio loss	1.44	1.43	1.38	1.36	1.66	1.76	1.50 ± 0.15
No feature learning	1.08	1.13	1.10	1.11	1.44	1.54	1.23 ± 0.18
No active selection	1.51	1.57	1.54	1.65	1.74	1.95	1.66 ± 0.15
No verification	0.61	0.61	0.61	0.62	0.63	0.63	0.62 ± 0.01
No weighted blending	1.12	1.39	1.33	1.46	1.92	2.03	1.54 ± 0.32
No CRF	1.38	1.40	1.39	1.54	1.84	1.96	1.58 ± 0.23
No verification stopping	1.67	1.69	1.65	1.69	1.95	2.05	1.78 ± 0.15