Few-shot classification on Mini-ImageNet 1-shot 5-way (test)
85.59AccuracyPTMAP-SF-SOTp
Evaluation Results
| Method | Links | |
|---|---|---|
| PTMAP-SF-SOTptransductive=true, backbone=wrn-28-102022.04 | 85.59 | |
| PTMAP-SFtransductive=true, backbone=wrn-28-102022.04 | 84.81 | |
| PTMAP-SOTttransductive=true, backbone=wrn-28-102022.04 | 84.18 | |
| Sill-Net-SOTptransductive=true, backbone=wrn-28-102022.04 | 83.35 | |
| PTMAP-SOTptransductive=true, backbone=wrn-28-102022.04 | 83.19 | |
| Sill-Nettransductive=true, backbone=wrn-28-102022.04 | 82.99 | |
| PTMAPtransductive=true, backbone=wrn-28-102022.04 | 82.92 | |
| PTMAP-SOFTMAXtransductive=true, backbone=wrn-28-102022.04 | 80.08 | |
| PTMAP-SOTptransductive=true, backbone=resnet-122022.04 | 78.35 | |
| PTMAP-SOTttransductive=true, backbone=resnet-122022.04 | 77.3 | |
| PTMAPtransductive=true, backbone=resnet-122022.04 | 76.9 | |
| PTMAP-COSINEtransductive=true, backbone=wrn-28-102022.04 | 74.6 | |
| ProtoNet-SOTttransductive=true, backbone=resnet-122022.04 | 67.9 | |
| RENettransductive=false, backbone=resnet-122022.04 | 67.6 | |
| ProtoNet-SOTptransductive=true, backbone=resnet-122022.04 | 67.34 | |
| FEATSetup=Embedding adaptation, Backbone=ResNet2018.12 | 66.78 | |
| FEATtransductive=false, backbone=resnet-122022.04 | 66.78 | |
| DeepEMDtransductive=false, backbone=resnet-122022.04 | 65.91 | |
| GCNSetup=Embedding adaptation, Backbone=ResNet2018.12 | 64.5 | |
| DEEPSETSSetup=Embedding adaptation, Backbone=ResNet2018.12 | 64.14 | |
| CTMBackbone=ResNet2018.12 | 64.12 | |
| BILSTMSetup=Embedding adaptation, Backbone=ResNet2018.12 | 63.9 | |
| SimpleShotBackbone=ResNet2018.12 | 62.85 | |
| MetaOptNetBackbone=ResNet2018.12 | 62.64 | |
| ProtoNettransductive=false, backbone=wrn-28-102022.04 | 62.6 | |
| ProtoNetSetup=Instance embedding, Backbone=ResNet2018.12 | 62.39 | |
| ProtoNettransductive=false, backbone=resnet-122022.04 | 62.39 | |
| PFABackbone=ResNet2018.12 | 59.6 | |
| TADAMBackbone=ResNet2018.12 | 58.5 | |
| FEAT‡Setup=Transductive, Backbone=ConvNet, Prediction Strategy=via Semi-ProtoNet2018.12 | 57.04 | |
| TEAMSetup=Transductive, Backbone=ConvNet2018.12 | 56.57 | |
| FEAT†Setup=Transductive, Backbone=ConvNet, Prediction Strategy=via ProtoNet2018.12 | 56.49 | |
| sparse-ReLU-MAMLBackbone=ResNet-12, Number of seeds=32021.10 | 56.39 | |
| TPNSetup=Transductive, Backbone=ConvNet2018.12 | 55.51 | |
| Semi-ProtoNet (Ours)Setup=Transductive, Backbone=ConvNet2018.12 | 55.5 | |
| ANILBackbone=ResNet-12, Number of seeds=32021.10 | 55.25 | |
| FEATSetup=Embedding adaptation, Backbone=ConvNet2018.12 | 55.15 | |
| FEATSetup=Standard, Backbone=ConvNet2018.12 | 55.15 | |
| FEATtransductive=false, backbone=conv-42022.04 | 55.15 | |
| sparse-MAMLBackbone=ResNet-12, Number of seeds=32021.10 | 55.02 | |
| PFABackbone=ConvNet2018.12 | 54.53 | |
| DEEPSETSSetup=Embedding adaptation, Backbone=ConvNet2018.12 | 54.41 | |
| ProtoNet-SOTptransductive=true, backbone=conv-42022.04 | 54.01 | |
| MAMLBackbone=ResNet-12, Number of seeds=32021.10 | 53.91 | |
| ProtoNet-SOTttransductive=true, backbone=conv-42022.04 | 53.7 | |
| GCNSetup=Embedding adaptation, Backbone=ConvNet2018.12 | 53.25 | |
| ProtoNetSetup=Instance embedding, Backbone=ConvNet2018.12 | 52.61 | |
| ProtoNetSetup=Standard, Backbone=ConvNet2018.12 | 52.61 | |
| MAML++Inner Steps=52018.10 | 52.15 | |
| MAML++Training Mode=Supervised Upper Bound2019.02 | 52.15 | |
| BILSTMSetup=Embedding adaptation, Backbone=ConvNet2018.12 | 52.13 | |
| MAML++Inner Steps=42018.10 | 51.65 | |
| MAML++Inner Steps=22018.10 | 51.49 | |
| RelationNetBackbone=ConvNet2018.12 | 51.38 | |
| MetaOptNetBackbone=ResNet-12, Number of seeds=3, Regularization=None2021.10 | 51.13 | |
| MAML++Inner Steps=32018.10 | 51.11 | |
| MAML++Inner Steps=12018.10 | 51.05 | |
| Meta-SGDInner Steps=12018.10 | 50.47 | |
| Semi-ProtoNetSetup=Transductive, Backbone=ConvNet2018.12 | 50.41 | |
| ProtoNetsTraining Mode=Supervised Upper Bound2019.02 | 50.16 | |
| ReptileTransduction=true2018.03 | 49.97 | |
| SimpleShotBackbone=ConvNet2018.12 | 49.69 | |
| ProtoNetBackbone=ConvNet2018.12 | 49.42 | |
| RelationNettransductive=false, backbone=conv-42022.04 | 49.31 | |
| Meta-Networks2018.10 | 49.21 | |
| ProtoNettransductive=false, backbone=conv-42022.04 | 49.1 | |
| MAMLTransduction=true2018.03 | 48.7 | |
| MAML (original paper)Inner Steps=52018.10 | 48.7 | |
| MAMLBackbone=ConvNet2018.12 | 48.7 | |
| MAML (local reproduction)Inner Steps=52018.10 | 48.25 | |
| 1st-order MAMLTransduction=true2018.03 | 48.07 | |
| ReptileTransduction=false2018.03 | 47.07 | |
| MAMLTraining Mode=Supervised Upper Bound2019.02 | 46.81 | |
| MAMLtransductive=false, backbone=conv-42022.04 | 46.47 | |
| Matching Nets2018.10 | 43.56 | |
| MatchNetBackbone=ConvNet2018.12 | 43.4 | |
| UMTRA + AutoAugmentLearning Paradigm=Unsupervised, Augmentation=AutoAugment2019.02 | 39.93 | |
| DeepCluster CACTUS-MAMLLearning Paradigm=Unsupervised, Clusterer=DeepCluster2019.02 | 39.9 | |
| DeepCluster CACTUS-ProtoNetsLearning Paradigm=Unsupervised, Clusterer=DeepCluster2019.02 | 39.18 | |
| AAL-ProtoNetsLearning Paradigm=Unsupervised, Algorithm=ProtoNets2019.02 | 37.67 | |
| AAL-MAML++Learning Paradigm=Unsupervised, Algorithm=MAML++2019.02 | 33.3 | |
| Training from scratchLearning Paradigm=Unsupervised2019.02 | 27.59 |