5-way Classification on miniImageNet (test)
70.1Accuracy (1-shot)LST
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LSTBackbone=ResNet-12 (pre), Strategy=recursive, hard, soft2019.06 | 70.1 | — | 78.7 | |
| Transductive fine-tuningArchitecture=WRN-28-10, Pre-training Split=train + val2019.09 | 68.11 | — | 80.36 | |
| Transductive fine-tuningArchitecture=WRN-28-10, Pre-training Split=train2019.09 | 65.73 | — | 78.4 | |
| MetaOpt SVMArchitecture=ResNet-12*, Pre-training Split=train + val2019.09 | 64.09 | — | 80 | |
| MetaOpt-SVMBackbone=ResNet-12, Method Category=Gradient descent, Note=Using 15-shot training samples on every meta-train task2019.06 | 62.64 | — | 78.63 | |
| MetaOpt SVMArchitecture=ResNet-12, Pre-training Split=train2019.09 | 62.64 | — | 78.63 | |
| LEOBackbone=WRN-28-10 (pre), Method Category=Gradient descent2019.06 | 61.76 | — | 77.59 | |
| LEOArchitecture=WRN-28-10, Pre-training Split=train + val2019.09 | 61.76 | — | 77.59 | |
| MTLBackbone=ResNet-12 (pre), Method Category=Gradient descent2019.06 | 61.2 | — | 75.5 | |
| Fine-tuningArchitecture=WRN-28-10, Pre-training Split=train + val2019.09 | 59.62 | — | 79.93 | |
| Activation to ParameterArchitecture=WRN-28-10, Pre-training Split=train + val2019.09 | 59.6 | — | 73.74 | |
| Transductive PropagationArchitecture=ResNet-122019.09 | 59.46 | — | 75.64 | |
| Delta-encoderBackbone=VGG-16 (pre), Method Category=Data augmentation2019.06 | 58.7 | — | 73.6 | |
| TADAMArchitecture=ResNet-122019.09 | 58.5 | — | 76.7 | |
| Support-based initializationArchitecture=WRN-28-10, Pre-training Split=train + val2019.09 | 58.47 | — | 75.56 | |
| Fine-tuningArchitecture=WRN-28-10, Pre-training Split=train2019.09 | 57.73 | — | 78.17 | |
| RCNBackbone=Res12, Type=Metric2020.09 | 57.4 | — | 75.19 | |
| adaResNetBackbone=ResNet-12 (One additional convolutional layer), Method Category=Gradient descent2019.06 | 56.88 | — | 71.94 | |
| AdaResNetBackbone=Res12, Type=Meta2020.09 | 56.88 | — | 71.94 | |
| Dynamic-NetBackbone=Conv4-64, Type=Meta2020.09 | 56.2 | — | 72.81 | |
| Support-based initializationArchitecture=WRN-28-10, Pre-training Split=train2019.09 | 56.17 | — | 73.31 | |
| SNAILBackbone=Res12, Type=Meta2020.09 | 55.71 | — | 68.88 | |
| Transductive PropagationArchitecture=conv (64)x42019.09 | 55.51 | — | 69.86 | |
| Dynamic-NetBackbone=Res12, Type=Meta2020.09 | 55.45 | — | 70.13 | |
| PARNBackbone=6-layer conv with deformable kernel, Type=Metric2020.09 | 55.22 | — | 71.55 | |
| Adv. ResNetBackbone=WRN-40 (pre), Method Category=Data augmentation2019.06 | 55.2 | — | 69.6 | |
| RCNBackbone=Conv4-64, Type=Metric2020.09 | 53.47 | — | 71.63 | |
| GCRBackbone=Conv4-512, Type=Metric2020.09 | 53.21 | — | 72.32 | |
| TPNBackbone=Conv4-64, Type=Metric2020.09 | 52.78 | — | 66.59 | |
| MetaGANBackbone=ResNet-12, Method Category=Gradient descent2019.06 | 52.71 | — | 68.63 | |
| PABNBackbone=Conv4-64, Type=Metric2020.09 | 51.87 | — | 65.37 | |
| R2-D2Backbone=Conv4-512, Type=Metric2020.09 | 51.8 | — | 68.4 | |
| R2D2Architecture=conv (96)x42019.09 | 51.8 | — | 68.4 | |
| DN4Backbone=Conv4-64, Type=Metric2020.09 | 51.24 | — | 71.02 | |
| Bilevel ProgrammingBackbone=ResNet-12 (Additional 2 convolutional layers), Method Category=Gradient descent2019.06 | 50.54 | — | 64.53 | |
| RELATION NETSBackbone=Conv4-64, Type=Metric2020.09 | 50.44 | — | 65.32 | |
| GNNBackbone=Conv4-64, Type=Metric2020.09 | 50.33 | — | 66.41 | |
| PROTOTYPICAL NETSBackbone=Conv4-64, Type=Metric2020.09 | 49.42 | — | 68.2 | |
| Prototypical NetworksArchitecture=conv (64)x42019.09 | 49.42 | — | 68.2 | |
| MAMLBackbone=4 CONV, Method Category=Gradient descent2019.06 | 48.7 | — | 63.11 | |
| MAMLBackbone=Conv4-32, Type=Meta2020.09 | 48.7 | — | 63.11 | |
| MAMLArchitecture=conv (32)x42019.09 | 48.7 | — | 63.11 | |
| Matching networksArchitecture=conv (64)x42019.09 | 46.6 | — | 60 | |
| Meta-LSTMBackbone=4 CONV, Method Category=Gradient descent2019.06 | 43.56 | — | 60.6 | |
| MATCHING NETSBackbone=Conv4-64, Type=Metric2020.09 | 43.56 | — | 55.31 | |
| META LSTMBackbone=Conv4-32, Type=Meta2020.09 | 43.44 | — | 60.6 | |
| LSTM meta-learnerArchitecture=conv (64)x42019.09 | 43.44 | — | 60.6 | |
| adaResNetFeature extractor=ResNet-12, Additional layers=1 convolutional layer2018.12 | — | 56.88 | — | |
| Adv. ResNetFeature extractor=WRN-40 (pre)2018.12 | — | 55.2 | — | |
| Bilevel ProgrammingFeature extractor=ResNet-12, Additional layers=2 convolutional layers2018.12 | — | 50.54 | — | |
| CompareNetsFeature extractor=4 CONV2018.12 | — | 50.44 | — | |
| Delta-encoderFeature extractor=VGG-16 (pre)2018.12 | — | 58.7 | — | |
| Hierarchical BayesFeature extractor=4 CONV2018.12 | — | 49.4 | — | |
| MAMLFeature extractor=4 CONV2018.12 | — | 48.7 | — | |
| MAML deep, HTFeature extractor=ResNet-12 (pre), Meta-batch type=HT meta-batch, Protocol=FT [Θ; θ]2018.12 | — | 59.1 | — | |
| MAML, HTFeature extractor=4 CONV, Meta-batch type=HT meta-batch, Protocol=FT [Θ; θ]2018.12 | — | 49.1 | — | |
| Matching NetsFeature extractor=4 CONV2018.12 | — | 43.44 | — | |
| Meta NetworksFeature extractor=5 CONV2018.12 | — | 49.21 | — | |
| Meta-LSTMFeature extractor=4 CONV2018.12 | — | 43.56 | — | |
| MetaGANFeature extractor=ResNet-122018.12 | — | 52.71 | — | |
| MTLFeature extractor=ResNet-12 (pre), Meta-batch type=meta-batch, Protocol=SS [Θ; θ]2018.12 | — | 60.2 | — | |
| MTLFeature extractor=ResNet-12 (pre), Meta-batch type=HT meta-batch, Protocol=SS [Θ; θ]2018.12 | — | 61.2 | — | |
| ProtoNetsFeature extractor=4 CONV2018.12 | — | 49.42 | — | |
| SNAILFeature extractor=ResNet-12 (pre)2018.12 | — | 55.71 | — | |
| TADAMFeature extractor=ResNet-12 (pre), Additional layers=72 fully connected layers2018.12 | — | 58.5 | — |