Image Classification on OMNIGLOT (test)
99.6AccuracyConvNet with Memory Module
Evaluation results
| Method | Links | |
|---|---|---|
| ConvNet with Memory ModuleN-way=5, K-shot=52017.03 | 99.6 | |
| Neural StatisticianK Shot=5, K Way=52016.06 | 99.5 | |
| MatchingK Shot=5, K Way=52016.06 | 98.9 | |
| Matching NetworkN-way=5, K-shot=52017.03 | 98.9 | |
| MatchingK Shot=5, K Way=202016.06 | 98.7 | |
| ConvNet with Memory ModuleN-way=20, K-shot=52017.03 | 98.6 | |
| Matching NetworkN-way=20, K-shot=52017.03 | 98.5 | |
| SiameseK Shot=5, K Way=52016.06 | 98.4 | |
| Convolutional Siamese NetN-way=5, K-shot=52017.03 | 98.4 | |
| ConvNet with Memory ModuleN-way=5, K-shot=12017.03 | 98.4 | |
| MatchingK Shot=1, K Way=52016.06 | 98.1 | |
| Neural StatisticianK Shot=1, K Way=52016.06 | 98.1 | |
| Neural StatisticianK Shot=5, K Way=202016.06 | 98.1 | |
| Matching NetworkN-way=5, K-shot=12017.03 | 98.1 | |
| SiameseK Shot=1, K Way=52016.06 | 97.3 | |
| SiameseK Shot=5, K Way=202016.06 | 97 | |
| Convolutional Siamese NetN-way=5, K-shot=12017.03 | 96.7 | |
| Convolutional Siamese NetN-way=20, K-shot=52017.03 | 96.5 | |
| ConvNet with Memory ModuleN-way=20, K-shot=12017.03 | 95 | |
| MANNK Shot=5, K Way=52016.06 | 94.9 | |
| MANN (no convolutions)N-way=5, K-shot=52017.03 | 94.9 | |
| MatchingK Shot=1, K Way=202016.06 | 93.8 | |
| Matching NetworkN-way=20, K-shot=12017.03 | 93.8 | |
| Neural StatisticianK Shot=1, K Way=202016.06 | 93.2 | |
| SiameseK Shot=1, K Way=202016.06 | 88.1 | |
| Convolutional Siamese NetN-way=20, K-shot=12017.03 | 88 | |
| VGG-B(S)Backbone=VGG-B, Pre-training=trained from scratch, #par=82017.05 | 86.9 | |
| DAN_sketchBackbone=VGG-B, Controller Initialization=sketch, #par=2.542017.05 | 85.4 | |
| DAN_imagenet+sketchBackbone=VGG-B, Controller Initialization=ImageNet + Sketch, #par=3.322017.05 | 85.4 | |
| VGG-B(P)Backbone=VGG-B, Pre-training=pre-trained on ImageNet, #par=82017.05 | 83.8 | |
| MANNK Shot=1, K Way=52016.06 | 82.8 | |
| MANN (no convolutions)N-way=5, K-shot=12017.03 | 82.8 | |
| DAN_imagenetBackbone=VGG-B, Controller Initialization=ImageNet, #par=2.762017.05 | 81.3 | |
| DAN_caltech-256Backbone=VGG-B, Controller Initialization=caltech-256, #par=2.542017.05 | 81 | |
| DAN_noiseBackbone=VGG-B, Controller Initialization=random weights, #par=1.762017.05 | 80.6 | |
| Pixels Nearest NeighborN-way=5, K-shot=52017.03 | 63.2 | |
| Pixels Nearest NeighborN-way=20, K-shot=52017.03 | 42.6 | |
| Pixels Nearest NeighborN-way=5, K-shot=12017.03 | 41.7 | |
| Pixels Nearest NeighborN-way=20, K-shot=12017.03 | 26.7 |