Few-shot classification on ImageNet Novel categories (test)
61.09Top-5 Acc (N'=1)SGM + Graph
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SGM + GraphBackbone=ResNet-502021.02 | 61.09 | 70.35 | 78.61 | — | — | |
| KGTNBackbone=ResNet-502021.02 | 60.1 | 69.4 | 78.1 | — | — | |
| KTCHBackbone=ResNet-502021.02 | 58.1 | 67.3 | 77.6 | — | — | |
| SGMBackbone=ResNet-502021.02 | 54.3 | 67 | 77.4 | — | — | |
| MatchingNetBackbone=ResNet-502021.02 | 53.5 | 63.5 | 72.7 | — | — | |
| PMNBackbone=ResNet-502021.02 | 53.3 | 65.2 | 75.9 | — | — | |
| ProtoNetBackbone=ResNet-502021.02 | 49.6 | 64 | 74.4 | — | — | |
| Cosine Classifier & Att. Weight GenBackbone=ResNet-10, Input resolution=224x224, Classifier type=Cosine similarity based ConvNet, Weight generator type=Attention-based, Dropout=0.52018.04 | 46.02 | 57.51 | 69.16 | 74.83 | 78.11 | |
| Prototype Matching Nets w/ HBackbone=ResNet-10, Input resolution=224x224, Hallucination (w/ H)=true2018.04 | 45.8 | 57.8 | 69 | 74.3 | 77.4 | |
| Cosine Classifier & Avg. Weight GenBackbone=ResNet-10, Input resolution=224x224, Classifier type=Cosine similarity based ConvNet, Weight generator type=Average weighting2018.04 | 45.23 | 56.9 | 68.68 | 74.36 | 77.69 | |
| Matching NetworksBackbone=ResNet-10, Input resolution=224x2242018.04 | 43.6 | 54 | 66 | 72.5 | 76.9 | |
| Prototype Matching NetsBackbone=ResNet-10, Input resolution=224x2242018.04 | 43.3 | 55.7 | 68.4 | 74 | 77 | |
| Logistic regression w/ HBackbone=ResNet-10, Input resolution=224x224, Hallucination (w/ H)=true2018.04 | 40.7 | 50.8 | 62 | 69.3 | 76.5 | |
| Prototypical-NetsBackbone=ResNet-10, Input resolution=224x2242018.04 | 39.3 | 54.4 | 66.3 | 71.2 | 73.9 | |
| Logistic regressionBackbone=ResNet-10, Input resolution=224x2242018.04 | 38.4 | 51.1 | 64.8 | 71.6 | 76.6 |