Research Article

Loss Architecture Search for Few-Shot Object Recognition

Table 4

Mean accuracy (%) ± CIs (%) of 14 state-of-the-art methods and LAS on the mini-ImageNet dataset with Conv-4 as the backbone.

Method5-way accuracy
1-Shot5-Shot

43.40 ± 0.7851.09 ± 0.71
MN-43.56 ± 0.8455.31 ± 0.73
ML-43.44 ± 0.7760.60 ± 0.71
45.1055.20
48.70 ± 1.8463.11 ± 0.92
Meta-49.21 ± 0.96
49.42 ± 0.7868.20 ± 0.66
49.02 ± 0.9863.50 ± 0.84
50.44 ± 0.8265.32 ± 0.70
Meta-50.47 ± 1.8764.03 ± 0.94
51.24 ± 0.7471.02 ± 0.64
51.77 ± 1.8665.60 ± 0.93
MM-Net52.74 ± 0.4565.82 ± 0.37
MM-Net53.37 ± 0.4866.97 ± 0.35
54.61 ± 0.8071.21 ± 0.66
LAS55.14 ± 0.4671.92 ± 0.48

Results from [32]. Results from [20]. Results from [33]. Results from [14].