Image Classification on ImageNet 1K Challenge (novel classes)
79.48Top-5 AccWDAE-GNN + EP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| WDAE-GNN + EPShots (K)=202020.03 | 79.48 | — | |
| CP-AANBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 79.3 | — | |
| Batch SGMShots (K)=202020.03 | 78.5 | — | |
| Cos & Att.Backbone=ResNet-10, m (number of training examples per class)=202019.05 | 78.1 | — | |
| LwoFShots (K)=202020.03 | 78.1 | — | |
| PMN w/ HBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 77.4 | — | |
| PMNShots (K)=202020.03 | 77.4 | — | |
| CC+ RotShots (K)=202020.03 | 77.31 | — | |
| WDAE-GNN + EPShots (K)=102020.03 | 77.25 | — | |
| PMNBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 77 | — | |
| CP-AANBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 76.5 | — | |
| Batch SGMShots (K)=102020.03 | 75.8 | — | |
| IDeMe-NetBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 75.1 | 45 | |
| WDAE-GNNShots (K)=202020.03 | 75 | — | |
| LRBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.8 | 44.9 | |
| Cos & Att.Backbone=ResNet-10, m (number of training examples per class)=102019.05 | 74.8 | — | |
| LwoFShots (K)=102020.03 | 74.8 | — | |
| Prototype ClassifierBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.7 | 44.1 | |
| CC+ RotShots (K)=102020.03 | 74.64 | — | |
| SVMBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.6 | 43.9 | |
| Generation SGMBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.6 | — | |
| FlippingBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.5 | 44.2 | |
| Gaussian NoiseBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.5 | 43.8 | |
| PMN w/ HBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 74.3 | — | |
| PMNShots (K)=102020.03 | 74.3 | — | |
| Gaussian Noise (feature level)Backbone=ResNet-10, m (number of training examples per class)=202019.05 | 74.2 | 44 | |
| PMNBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 74 | — | |
| IDeMe-NetBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 73.4 | 42.7 | |
| MixupBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 72.9 | 42.1 | |
| WDAE-GNNShots (K)=102020.03 | 72.89 | — | |
| WDAE-GNN + EPShots (K)=52020.03 | 72.89 | — | |
| Matching NetworkBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 72.8 | — | |
| Prototypical NetworkBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 72.3 | 42.7 | |
| SoftmaxBackbone=ResNet-10, m (number of training examples per class)=202019.05 | 72.1 | — | |
| Batch SGMShots (K)=52020.03 | 71.4 | — | |
| LRBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 71.3 | 41.1 | |
| Generation SGMBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 70.5 | — | |
| IDeMe-NetBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 70.4 | 39.3 | |
| WDAE-GNNShots (K)=52020.03 | 70.3 | — | |
| CP-AANBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 70.2 | — | |
| FlippingBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 70.2 | 38.7 | |
| Prototype ClassifierBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 69.9 | 38.4 | |
| Gaussian NoiseBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 69.7 | 38 | |
| CC+ RotShots (K)=52020.03 | 69.67 | — | |
| Gaussian Noise (feature level)Backbone=ResNet-10, m (number of training examples per class)=102019.05 | 69.5 | 38.2 | |
| SVMBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 69.2 | 37.9 | |
| MixupBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 69.2 | 38.5 | |
| LwoFShots (K)=52020.03 | 69.2 | — | |
| Cos & Att.Backbone=ResNet-10, m (number of training examples per class)=52019.05 | 69.1 | — | |
| PMN w/ HBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 69 | — | |
| PMNShots (K)=52020.03 | 69 | — | |
| Matching NetworkBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 68.5 | — | |
| PMNBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 68.4 | — | |
| Prototypical NetworkBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 68.2 | 37.7 | |
| SoftmaxBackbone=ResNet-10, m (number of training examples per class)=102019.05 | 67.3 | — | |
| LRBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 66.1 | 35.8 | |
| Matching NetworkBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 64.4 | — | |
| Generation SGMBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 64.1 | — | |
| FlippingBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 64.1 | 33.7 | |
| Prototype ClassifierBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 63.9 | 33.8 | |
| Prototypical NetworkBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 63.7 | 33.5 | |
| Gaussian NoiseBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 63.7 | 33.9 | |
| Gaussian Noise (feature level)Backbone=ResNet-10, m (number of training examples per class)=52019.05 | 63.3 | 33.4 | |
| WDAE-GNN + EPShots (K)=22020.03 | 62.16 | — | |
| SVMBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 61.2 | 31.5 | |
| MixupBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 61.1 | 32 | |
| IDeMe-NetBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 60.9 | 30.1 | |
| Batch SGMShots (K)=22020.03 | 60.5 | — | |
| WDAE-GNNShots (K)=22020.03 | 59.7 | — | |
| CP-AANBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 59.3 | — | |
| PMN w/ HBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 57.8 | — | |
| PMNShots (K)=22020.03 | 57.8 | — | |
| CC+ RotShots (K)=22020.03 | 57.8 | — | |
| Cos & Att.Backbone=ResNet-10, m (number of training examples per class)=22019.05 | 57.5 | — | |
| LwoFShots (K)=22020.03 | 57.5 | — | |
| SoftmaxBackbone=ResNet-10, m (number of training examples per class)=52019.05 | 57.4 | — | |
| PMNBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 55.7 | — | |
| LRBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 54.7 | 26 | |
| Matching NetworkBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 54.1 | — | |
| Prototypical NetworkBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 53.6 | 24 | |
| Gaussian Noise (feature level)Backbone=ResNet-10, m (number of training examples per class)=22019.05 | 51.4 | 24.2 | |
| MixupBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 51.4 | 24.6 | |
| FlippingBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 51.2 | 24.7 | |
| Gaussian NoiseBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 51.2 | 24 | |
| Prototype ClassifierBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 51.1 | 24.3 | |
| IDeMe-NetBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 51 | 23.1 | |
| WDAE-GNN + EPShots (K)=12020.03 | 50.07 | — | |
| Batch SGMShots (K)=12020.03 | 49.3 | — | |
| Generation SGMBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 48.9 | — | |
| SVMBackbone=ResNet-10, m (number of training examples per class)=22019.05 | 48.4 | 22.7 | |
| CP-AANBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 48.4 | — | |
| WDAE-GNNShots (K)=12020.03 | 48 | — | |
| CC+ RotShots (K)=12020.03 | 46.43 | — | |
| LwoFShots (K)=12020.03 | 46.2 | — | |
| Cos & Att.Backbone=ResNet-10, m (number of training examples per class)=12019.05 | 46 | — | |
| PMN w/ HBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 45.8 | — | |
| PMNShots (K)=12020.03 | 45.8 | — | |
| PMNBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 43.3 | — | |
| Matching NetworkBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 43 | — | |
| LRBackbone=ResNet-10, m (number of training examples per class)=12019.05 | 42.8 | 18.3 |