Few-shot Image Classification on Mini-Imagenet (test)
99AccuracyAdult (human)
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Adult (human)shot=1-shot, scenario=5-way2019.06 | 99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M-SVM (best) (our)shot=5-shot2020.02 | 82.13 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Multiple-Semanticsshot=5-shot, scenario=5-way, semantics=labels and descriptions2019.06 | 82.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| R2-D2 (best) (our)shot=5-shot2020.02 | 81.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BD-CSPNSetting=Transductive, Backbone=WRN-28-10, Shots=5-shot2019.11 | 81.89 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SimpleShotshot=5-shot, scenario=5-way, semantics=none2019.06 | 81.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MC2shots=5-shot, feature extraction=multiview, meta-training protocol=meta-train + meta-validation, backbone=fixed features from [2]2019.02 | 80.21 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNet-SVMshots=5-shot, 15-shot meta-training=true, meta-training protocol=meta-train + meta-validation, backbone=ResNet-122019.02 | 80 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M-SVMshot=5-shot2020.02 | 80 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BFSLSetting=Inductive, Backbone=WRN-28-10, Shots=5-shot2019.11 | 79.87 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOshots=5-shot, feature extraction=multiview, backbone=fixed features from [2]2019.02 | 79.49 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetAdaptshot=5-shot, scenario=5-way, semantics=none2019.06 | 79.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSTSetting=Semi-Supervised, Backbone=ResNet-12, Shots=5-shot2019.11 | 78.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNet-SVMSetting=Inductive, Backbone=ResNet-12, Shots=5-shot2019.11 | 78.63 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNetshot=5-shot, scenario=5-way, semantics=none2019.06 | 78.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Transductive Fine-TuningSetting=Transductive, Backbone=WRN-28-10, Shots=5-shot2019.11 | 78.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AM3-TADAMshot=5-shot, scenario=5-way, semantics=labels and descriptions2019.06 | 78.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOshot=5-shot, scenario=5-way, semantics=none2019.06 | 77.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOshots=5-shot, feature extraction=center, backbone=fixed features from [2]2019.02 | 77.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOshot=5-shot2020.02 | 77.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOEmbedding Net=Wide-ResNet282019.04 | 77.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOSetting=Inductive, Backbone=WRN-28-10, Shots=5-shot, Training set and validation set used for training=true2019.11 | 77.59 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MC2shots=5-shot, feature extraction=center, meta-training protocol=meta-train + meta-validation, backbone=fixed features from [2]2019.02 | 77.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dist. ensembleshot=5-shot, scenario=5-way, semantics=none2019.06 | 76.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TriNetshot=5-shot, scenario=5-way, semantics=labels and descriptions2019.06 | 76.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hyperbolic ProtoNetEmbedding Net=ResNet182019.04 | 76.84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Self-Jig(SVM)Embedding Net=ResNet502019.04 | 76.71 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TADAMshot=5-shot2020.02 | 76.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TADAMEmbedding Net=ResNet122019.04 | 76.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TADAMSetting=Inductive, Backbone=ResNet-12, Shots=5-shot2019.11 | 76.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EGNN+TransductionTransduction=Yes, Way=5-way, Shot=5-shot2019.05 | 76.37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EGNNSetting=Transductive, Backbone=ConvNet-256, Shots=5-shot2019.11 | 76.37 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MC2shots=5-shot, feature extraction=center, backbone=fixed features from [2]2019.02 | 75.92 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPNEmbedding Net=ResNet122019.04 | 75.65 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNets*shot=5-shot2020.02 | 75.6 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MTLEmbedding Net=ResNet122019.04 | 75.5 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DN4Embedding Net=ResNet122019.04 | 74.44 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNet+Embedding Net=ResNet122019.04 | 74.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Dis. k-shotEmbedding Net=ResNet342019.04 | 73.9 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| [33]shots=5-shot, 15-shot meta-training=true, backbone=fixed features from [2]2019.02 | 73.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Qiao-WRNEmbedding Net=Wide-ResNet282019.04 | 73.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Activation to ParameterSetting=Inductive, Backbone=WRN-28-10, Shots=5-shot, Training set and validation set used for training=true2019.11 | 73.74 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| [7]shot=5-shot2020.02 | 73 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LwoFSetting=Inductive, Backbone=ConvNet-128, Shots=5-shot2019.11 | 73 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Hyperbolic ProtoNetEmbedding Net=4 Conv2019.04 | 72.67 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAMLshot=5-shot, scenario=5-way, semantics=none2019.06 | 72.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| CAMLEmbedding Net=ResNet122019.04 | 72.35 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| adaResNetShot=5-shot, Conditioning=direct feedback (DF)2017.12 | 71.94 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AdaResNetSetting=Inductive, Backbone=ResNet-12, Shots=5-shot2019.11 | 71.94 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| adaResNetShot=5-shot, Conditioning=error gradient (V)2017.12 | 71.69 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DEML+Meta-SGDshot=5-shot, scenario=5-way, semantics=none2019.06 | 71.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| DN4Embedding Net=4 Conv2019.04 | 71.02 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| BD-CSPNSetting=Transductive, Backbone=WRN-28-10, Shots=1-shot2019.11 | 70.31 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LSTSetting=Semi-Supervised, Backbone=ResNet-12, Shots=1-shot2019.11 | 70.1 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 4.5 years old (human)shot=1-shot, scenario=5-way2019.06 | 70 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPN (Higher Shot)Transduction=Yes, Way=5-way, Shot=5-shot2018.05 | 69.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPN (Higher K)Transduction=Yes, Way=5-way, Shot=5-shot2019.05 | 69.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPNSetting=Transductive, Backbone=ConvNet-64, Shots=5-shot2019.11 | 69.86 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Delta-encodershot=5-shot, scenario=5-way, semantics=none2019.06 | 69.7 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPNTransduction=Yes, Way=5-way, Shot=5-shot2018.05 | 69.43 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TPNTransduction=Yes, Way=5-way, Shot=5-shot2019.05 | 69.43 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| TCMLShot=5-shot2017.12 | 68.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNAILEmbedding Net=ResNet122019.04 | 68.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SNAILSetting=Inductive, Backbone=ResNet-12, Shots=5-shot2019.11 | 68.88 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Meta-SGDshots=5-shot, feature extraction=center, backbone=fixed features from [2]2019.02 | 68.84 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PROTO NET (Higher Way)Transduction=No, Way=5-way, Shot=5-shot2018.05 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ProtoNetEmbedding Net=4 Conv2019.04 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Prototypical NetworksSetting=Inductive, Backbone=ConvNet-64, Shots=5-shot, Training set and validation set used for training=true2019.11 | 68.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Label PropagationTransduction=Yes, Way=5-way, Shot=5-shot2018.05 | 68.18 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Spot&learnEmbedding Net=4 Conv2019.04 | 67.96 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Multiple-Semanticsshot=1-shot, scenario=5-way, semantics=labels and descriptions2019.06 | 67.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RELATION NETTransduction=BN, Way=5-way, Shot=5-shot2018.05 | 67.07 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Relation NetTransduction=BN, Way=5-way, Shot=5-shot2019.05 | 67.07 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| EGNNTransduction=No, Way=5-way, Shot=5-shot2019.05 | 66.85 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| GNNTransduction=No, Way=5-way, Shot=5-shot2019.05 | 66.41 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAML+TransductionTransduction=Yes, Way=5-way, Shot=5-shot2018.05 | 66.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAML+TransductionTransduction=Yes, Way=5-way, Shot=5-shot2019.05 | 66.19 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| ReptileTransduction=true2018.03 | 65.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Reptile + BNTransduction=BN, Way=5-way, Shot=5-shot2018.05 | 65.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Reptile + BNTransduction=BN, Way=5-way, Shot=5-shot2019.05 | 65.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| REPTILEEmbedding Net=4 Conv2019.04 | 65.99 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| R2-D2 (best) (our)shot=1-shot2020.02 | 65.95 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| PROTO NETTransduction=No, Way=5-way, Shot=5-shot2018.05 | 65.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Prototypical NetTransduction=No, Way=5-way, Shot=5-shot2019.05 | 65.77 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Transductive Fine-TuningSetting=Transductive, Backbone=WRN-28-10, Shots=1-shot2019.11 | 65.73 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M-SVM (best) (our)shot=1-shot2020.02 | 65.38 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| RelationNetEmbedding Net=4 Conv2019.04 | 65.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| Relation NetSetting=Inductive, Backbone=ConvNet-256, Shots=5-shot2019.11 | 65.32 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| AM3-TADAMshot=1-shot, scenario=5-way, semantics=labels and descriptions2019.06 | 65.3 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MC2shots=1-shot, feature extraction=multiview, meta-training protocol=meta-train + meta-validation, backbone=fixed features from [2]2019.02 | 64.4 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| SimpleShotshot=1-shot, scenario=5-way, semantics=none2019.06 | 64.2 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MetaOptNet-SVMshots=1-shot, 15-shot meta-training=true, meta-training protocol=meta-train + meta-validation, backbone=ResNet-122019.02 | 64.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| M-SVMshot=1-shot2020.02 | 64.09 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| LEOshots=1-shot, feature extraction=multiview, backbone=fixed features from [2]2019.02 | 63.97 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| 1st-order MAMLTransduction=true2018.03 | 63.15 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAMLTransduction=true2018.03 | 63.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAMLTransduction=BN, Way=5-way, Shot=5-shot2018.05 | 63.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAMLTransduction=BN, Way=5-way, Shot=5-shot2019.05 | 63.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAMLEmbedding Net=4 Conv2019.04 | 63.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | |
| MAMLSetting=Inductive, Backbone=ConvNet-32, Shots=5-shot2019.11 | 63.11 | — | — | — | — | — | — | — | — | — | — | — | — | — | — |