Few-shot Classification on i-Nat (test)
84.3AccuracyPADDLE
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PADDLEBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 84.3 | — | — | — | |
| LaplacianShotNetwork=WRN, Feature Normalization=UN2020.06 | 74.97 | 71.55 | — | — | |
| LaplacianShotNetwork=ResNet-50, Feature Normalization=UN2020.06 | 69.13 | 65.96 | — | — | |
| LaplacianShotNetwork=WRN, Feature Normalization=L22020.06 | 67.82 | 65.78 | — | — | |
| LaplacianShotNetwork=WRN, Feature Normalization=CL22020.06 | 67.43 | 65.32 | — | — | |
| alpha-TIMBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 66.7 | — | — | — | |
| LaplacianShotNetwork=ResNet-18, Feature Normalization=UN2020.06 | 66.4 | 62.8 | — | — | |
| SimpleShotNetwork=WRN, Feature Normalization=L22020.06 | 66.25 | 64.26 | — | — | |
| SimpleShotNetwork=WRN, Feature Normalization=CL22020.06 | 65.17 | 63.03 | — | — | |
| SimpleShotNetwork=WRN, Feature Normalization=UN2020.06 | 65.08 | 62.44 | — | — | |
| LaplacianShotNetwork=ResNet-50, Feature Normalization=L22020.06 | 63.66 | 61.4 | — | — | |
| LaplacianShotNetwork=ResNet-50, Feature Normalization=CL22020.06 | 63.18 | 61.08 | — | — | |
| SimpleShotNetwork=ResNet-50, Feature Normalization=L22020.06 | 61.99 | 59.68 | — | — | |
| LaplacianShotNetwork=ResNet-18, Feature Normalization=L22020.06 | 61.14 | 58.72 | — | — | |
| SimpleShotNetwork=ResNet-50, Feature Normalization=UN2020.06 | 61.07 | 58.45 | — | — | |
| SimpleShotNetwork=ResNet-50, Feature Normalization=CL22020.06 | 60.98 | 58.83 | — | — | |
| LaplacianShotNetwork=ResNet-18, Feature Normalization=CL22020.06 | 60.81 | 58.49 | — | — | |
| SimpleShotNetwork=ResNet-18, Feature Normalization=L22020.06 | 59.56 | 57.15 | — | — | |
| SimpleShotNetwork=ResNet-18, Feature Normalization=CL22020.06 | 58.63 | 56.35 | — | — | |
| SimpleShotNetwork=ResNet-18, Feature Normalization=UN2020.06 | 58.56 | 55.8 | — | — | |
| BaselineBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 58.2 | — | — | — | |
| BD-CSPNBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 57.6 | — | — | — | |
| Wertheimer & HariharanNetwork=ResNet-502020.06 | 51.25 | 46.04 | — | — | |
| LaplacianShotBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 43.3 | — | — | — | |
| TIMBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 37.5 | — | — | — | |
| PT-MAPBackbone=ResNet-18, Number of ways=227, Effective classes (Keff)=52022.10 | 6.8 | — | — | — | |
| Rot + KD + CEnegative_sample_setup=noisy negative samples2022.06 | — | — | 48.9 | 66.58 | |
| Rot + KD + CEnegative_sample_setup=disjoint negative samples2022.06 | — | — | 48.8 | 66.72 | |
| Rot + KD + POODLE-Bnegative_sample_setup=noisy negative samples2022.06 | — | — | 50.62 | 67.31 | |
| Rot + KD + POODLE-Bnegative_sample_setup=disjoint negative samples2022.06 | — | — | 50.46 | 67.45 | |
| Rot + KD + POODLE-Rnegative_sample_setup=noisy negative samples2022.06 | — | — | 48.94 | 66.87 | |
| Rot + KD + POODLE-Rnegative_sample_setup=disjoint negative samples2022.06 | — | — | 48.84 | 67.03 |