Low-shot Image Classification on ImageNet base and novel classes (test val)
63.6Top-5 Acc (n=1)SGM + Generation
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SGM + GenerationBackbone=ResNet-50, Lowshot strategy=Generation + Classifier2016.06 | 63.6 | 71.5 | 80 | 83.3 | 85.2 | |
| SGM* + Generation*Backbone=ResNet-50, Lowshot phase=Generation* + Classifier2016.06 | 63.6 | 71.5 | 80 | 83.3 | 85.2 | |
| Baseline + GenerationBackbone=ResNet-50, Lowshot strategy=Generation + Classifier2016.06 | 63.1 | 71.5 | 78.8 | 82.6 | 85.4 | |
| SGMBackbone=ResNet-50, Lowshot strategy=Classifier2016.06 | 60 | 71.3 | 80 | 83.3 | 85.2 | |
| Matching NetworkBackbone=ResNet-10, Lowshot strategy=Matching Network2016.06 | 55 | 61.5 | 69.3 | 73.4 | 76.2 | |
| SGM + GenerationBackbone=ResNet-10, Lowshot strategy=Generation + Classifier2016.06 | 54.3 | 62.1 | 71.3 | 75.8 | 78.1 | |
| SGM* + Generation*Backbone=ResNet-10, Lowshot phase=Generation* + Classifier2016.06 | 54.3 | 62.1 | 71.3 | 75.8 | 78.1 | |
| BaselineBackbone=ResNet-50, Lowshot strategy=Classifier2016.06 | 54.1 | 67.7 | 79.1 | 83.2 | 85.4 | |
| BaselineBackbone=ResNet-50, Lowshot phase=Classifier2016.06 | 54.1 | 67.7 | 79.1 | 83.2 | 85.4 | |
| L2Backbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 52.7 | 63 | 71.5 | 74.8 | 76.4 | |
| Baseline + GenerationBackbone=ResNet-10, Lowshot strategy=Generation + Classifier2016.06 | 52.4 | 59.4 | 67.5 | 72.6 | 76.9 | |
| Baseline + Generation*Backbone=ResNet-10, Lowshot phase=Generation* + Classifier2016.06 | 52.4 | 59.4 | 67.5 | 72.6 | 76.9 | |
| DropoutBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 50.1 | 59.7 | 68.8 | 72.7 | 74.7 | |
| SGMBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 49.4 | 60.5 | 71.3 | 75.8 | 78.1 | |
| SGM*Backbone=ResNet-10, Lowshot phase=Classifier2016.06 | 49.4 | 60.5 | 71.3 | 75.8 | 78.1 | |
| Batch SGMBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 49.3 | 60.5 | 71.4 | 75.8 | 78.5 | |
| Batch SGM*Backbone=ResNet-10, Lowshot phase=Classifier2016.06 | 49.3 | 60.5 | 71.4 | 75.8 | 78.5 | |
| TripletsBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 47.6 | 57.1 | 65.2 | 68.4 | 70.2 | |
| L1Backbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 47.1 | 58.5 | 69.2 | 73.7 | 76.1 | |
| Model RegressionBackbone=ResNet-10, Lowshot strategy=Model Regression2016.06 | 46.4 | 56.7 | 66.8 | 70.4 | 72 | |
| Baseline (Data Augmentation)Backbone=ResNet-10, Lowshot strategy=Data Augmentation2016.06 | 44.9 | 54 | 66.4 | 73 | 77.2 | |
| MultiverseBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 44.1 | 54.2 | 67 | 73.2 | 76.9 | |
| DecovBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 43.3 | 55.7 | 70.1 | 75.4 | 77.9 | |
| BaselineBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 43 | 54.3 | 67.2 | 72.8 | 75.9 | |
| BaselineBackbone=ResNet-10, Lowshot phase=Classifier2016.06 | 43 | 54.3 | 67.2 | 72.8 | 75.9 | |
| Baseline-ftBackbone=ResNet-10, Lowshot strategy=Classifier2016.06 | 41.7 | 51.7 | 65 | 71.2 | 74.5 |