Table 6 Zero-shot recognition results (± standard deviation) of the SAEVRT method compared with those of SOTA methods (under 14/7, 16/5 and 18/3 splits) for the UCM21 dataset (\(\%\)).
From: Enhancing zero-shot scene recognition through semantic autoencoders and visual relation transfer
Method | 14/7 | 16/5 | 18/3 |
|---|---|---|---|
SSE | 20.87 ± 6.85 | 35.59 ± 5.90 | 49.38 ± 9.91 |
RKT | 32.15 ± 7.07 | 45.26 ± 7.83 | 58.45 ± 10.67 |
DMaP | 36.41 ± 7.96 | 48.92 ± 8.71 | 62.33 ± 9.86 |
SAE | 37.94 ± 8.75 | 49.50 ± 8.42 | 64.02 ± 10.90 |
WDVSc | 40.28 ± 10.12 | 55.91 ± 11.77 | 66.69 ± 11.14 |
VSOP | 40.21 ± 9.71 | 46.48 ± 7.83 | 66.72 ± 12.37 |
LsrGAN | 41.40 ± 10.54 | 53.71 ± 9.03 | 68.44 ± 12.45 |
SAGAN | 42.56 ± 10.68 | 55.65 ± 10.67 | 67.31 ± 13.13 |
Ma et al. | 43.38 ± 10.05 | 56.14 ± 10.06 | 69.00 ± 13.55 |
Liu et al. | 43.63 ± 9.76 | 57.77 ± 10.50 | 68.83 ± 13.89 |
SAEVRT | 44.69 ± 10.65 | 58.68 ± 10.04 | 70.55 ± 13.32 |
