Generalized Few-Shot Learning on ImageNet 2012 (Novel classes)
79.4Top-5 AccuracyDynamic FSL + TRAML
Evaluation Results
| Method | Links | |
|---|---|---|
| Dynamic FSL + TRAMLNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 79.4 | |
| Dynamic FSLNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 78.1 | |
| Prototype Matching Nets w/HNumber of shots (ns)=20, Hallucination=true, Backbone=ResNet102020.05 | 77.4 | |
| Prototype Matching NetsNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 77 | |
| Matching NetworksNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 76.9 | |
| Logistic regressionNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 76.6 | |
| Logistic regression w/HNumber of shots (ns)=20, Hallucination=true, Backbone=ResNet102020.05 | 76.5 | |
| Dynamic FSL + TRAMLNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 76.4 | |
| Dynamic FSLNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 74.8 | |
| Prototype Matching Nets w/HNumber of shots (ns)=10, Hallucination=true, Backbone=ResNet102020.05 | 74.3 | |
| Prototype Matching NetsNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 74 | |
| Prototypical NetworkNumber of shots (ns)=20, Hallucination=false, Backbone=ResNet102020.05 | 73.9 | |
| Matching NetworksNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 72.5 | |
| Logistic regressionNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 71.6 | |
| Prototypical NetworkNumber of shots (ns)=10, Hallucination=false, Backbone=ResNet102020.05 | 71.2 | |
| Dynamic FSL + TRAMLNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 70.3 | |
| Logistic regression w/HNumber of shots (ns)=10, Hallucination=true, Backbone=ResNet102020.05 | 69.3 | |
| Dynamic FSLNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 69.2 | |
| Prototype Matching Nets w/HNumber of shots (ns)=5, Hallucination=true, Backbone=ResNet102020.05 | 69 | |
| Prototype Matching NetsNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 68.4 | |
| Prototypical NetworkNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 66.3 | |
| Matching NetworksNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 66 | |
| Logistic regressionNumber of shots (ns)=5, Hallucination=false, Backbone=ResNet102020.05 | 64.8 | |
| Logistic regression w/HNumber of shots (ns)=5, Hallucination=true, Backbone=ResNet102020.05 | 62 | |
| Knowledge Graph Transfer Networkk-shot=52019.11 | 60.7 | |
| KTCHk-shot=52019.11 | 60.5 | |
| Dynamic FSL + TRAMLNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 59.2 | |
| KTCHk-shot=42019.11 | 58.7 | |
| Knowledge Graph Transfer Networkk-shot=42019.11 | 58.4 | |
| Prototype Matching Nets w/HNumber of shots (ns)=2, Hallucination=true, Backbone=ResNet102020.05 | 57.8 | |
| LSDk-shot=52019.11 | 57.6 | |
| Dynamic FSLNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 57.5 | |
| SGMk-shot=52019.11 | 56.8 | |
| Prototype Matching NetsNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 55.7 | |
| PPAk-shot=52019.11 | 55.4 | |
| Knowledge Graph Transfer Networkk-shot=32019.11 | 55.4 | |
| KTCHk-shot=32019.11 | 54.9 | |
| Prototypical NetworkNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 54.4 | |
| Matching NetworksNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 54 | |
| NNk-shot=52019.11 | 54 | |
| SGMk-shot=42019.11 | 53.5 | |
| LSDk-shot=42019.11 | 53.4 | |
| PPAk-shot=42019.11 | 52.5 | |
| NNk-shot=42019.11 | 52.3 | |
| Logistic regressionNumber of shots (ns)=2, Hallucination=false, Backbone=ResNet102020.05 | 51.1 | |
| Logistic regression w/HNumber of shots (ns)=2, Hallucination=true, Backbone=ResNet102020.05 | 50.8 | |
| Knowledge Graph Transfer Networkk-shot=22019.11 | 50.3 | |
| LSDk-shot=32019.11 | 50.2 | |
| SGMk-shot=32019.11 | 49 | |
| KTCHk-shot=22019.11 | 48.9 | |
| NNk-shot=32019.11 | 48.7 | |
| PPAk-shot=32019.11 | 48.5 | |
| Dynamic FSL + TRAMLNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 48.1 | |
| Dynamic FSLNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 46 | |
| Prototype Matching Nets w/HNumber of shots (ns)=1, Hallucination=true, Backbone=ResNet102020.05 | 45.8 | |
| LSDk-shot=22019.11 | 44.7 | |
| Matching NetworksNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 43.6 | |
| NNk-shot=22019.11 | 43.6 | |
| Prototype Matching NetsNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 43.3 | |
| PPAk-shot=22019.11 | 43.1 | |
| SGMk-shot=22019.11 | 42.5 | |
| Knowledge Graph Transfer Networkk-shot=12019.11 | 42.5 | |
| Logistic regression w/HNumber of shots (ns)=1, Hallucination=true, Backbone=ResNet102020.05 | 40.7 | |
| Prototypical NetworkNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 39.3 | |
| KTCHk-shot=12019.11 | 39 | |
| Logistic regressionNumber of shots (ns)=1, Hallucination=false, Backbone=ResNet102020.05 | 38.4 | |
| NNk-shot=12019.11 | 34.2 | |
| LSDk-shot=12019.11 | 33.2 | |
| PPAk-shot=12019.11 | 33 | |
| SGMk-shot=12019.11 | 31.6 |