Table 8.

Comparison of different variants of our method on VOC2007. 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	0.59	0.68	0.73	0.63	0.90	1.12	0.77 ± 0.19
No log-ratio loss	0.30	0.35	0.39	0.30	0.50	0.71	0.42 ± 0.14
No feature learning	0.27	0.30	0.33	0.28	0.40	0.54	0.35 ± 0.09
No active selection	0.58	0.66	0.69	0.62	0.85	1.01	0.73 ± 0.15
No verification	0.45	0.47	0.47	0.45	0.48	0.53	0.47 ± 0.03
No weighted blending	0.32	0.43	0.48	0.35	0.81	0.99	0.56 ± 0.25
No CRF	0.56	0.63	0.69	0.60	0.89	1.09	0.74 ± 0.19
No verification stopping	0.53	0.65	0.62	0.55	0.89	1.09	0.72 ± 0.20