Table 4.

Quantitative results on UCM multi-label dataset (%).

Model	m.$F_1$	m.$F_2$	m.${\mathrm{P}}_e$	m.${\mathrm{R}}_e$	m.${\mathrm{P}}_l$	m.${\mathrm{R}}_l$
VGGNet (Simonyan and Zisserman, 2014)	78.54	80.17	79.06	82.30	86.02	80.21
VGG-RBFNN (Zeggada et al., 2017)	78.80	81.14	78.18	83.91	81.90	82.63
CA-VGG-LSTM	79.57	80.75	80.64	82.47	87.74	75.95
CA-VGG-BiLSTM	79.78	81.69	79.33	83.99	85.28	76.52
GoogLeNet (Szegedy et al., 2015)	80.68	82.32	80.51	84.27	87.51	80.85
GoogLeNet-RBFNN (Zeggada et al., 2017)	81.54	84.05	79.95	86.75	86.19	84.92
CA-GoogLeNet-LSTM	81.78	85.16	78.52	88.60	86.66	85.99
CA-GoogLeNet-BiLSTM	81.82	84.41	79.91	87.06	86.29	84.38
ResNet-50 (He et al., 2016)	79.68	80.58	80.86	81.95	88.78	78.98
ResNet-RBFNN (Zeggada et al., 2017)	80.58	82.47	79.92	84.59	86.21	83.72
CA-ResNet-LSTM	81.36	83.66	79.90	86.14	86.99	82.24
CA-ResNet-BiLSTM	81.47	85.27	77.94	89.02	86.12	84.26

m.$F_1$ and m.$F_2$ indicate the mean $F_1$ and $F_2$ score.

m.${\mathrm{P}}_e$ and m.${\mathrm{R}}_e$ indicate mean example-based precision and recall.

m.${\mathrm{P}}_l$ and m.${\mathrm{R}}_l$ indicate mean label-based precision and recall.