Table 7 Zero-shot recognition results (± standard deviation) of the SAEVRT method compared with those of SOTA methods (under 30/15, 35/10 and 40/5 splits) for the NWPU45 dataset (\(\%\)).
From: Enhancing zero-shot scene recognition through semantic autoencoders and visual relation transfer
Method | 30/15 | 35/10 | 40/5 |
|---|---|---|---|
SSE | 23.30 ± 2.48 | 33.36 ± 3.58 | 44.56 ± 6.13 |
RKT | 34.18 ± 3.41 | 43.72 ± 4.30 | 53.36 ± 6.64 |
DMaP | 38.07 ± 4.83 | 49.53 ± 6.31 | 57.17 ± 6.85 |
SAE | 35.07 ± 3.91 | 44.81 ± 4.73 | 55.42 ± 7.49 |
WDVSc | 40.92 ± 4.59 | 50.68 ± 6.60 | 60.73 ± 7.18 |
VSOP | 36.09 ± 4.63 | 45.32 ± 5.71 | 59.49 ± 7.38 |
LsrGAN | 37.69 ± 4.54 | 50.26 ± 5.75 | 62.31 ± 8.40 |
SAGAN | 38.50 ± 4.40 | 51.51 ± 6.10 | 62.66 ± 8.53 |
Ma et al. | 40.33 ± 3.71 | 52.23 ± 6.48 | 64.15 ± 8.81 |
Liu et al. | 41.06 ± 4.38 | 52.47 ± 6.50 | 63.90 ± 9.45 |
SAEVRT | 42.56 ± 3.45 | 53.26 ± 6.11 | 65.70 ± 8.37 |
