Few-shot Classification on miniImageNet (N-way Accuracy)
98.45-way 5-shot AccuracyP>M>F
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| P>M>FVision Backbone=ViT-B/16, External Data=true2022.11 | 98.4 | — | — | — | 95.3 | — | — | — | |
| CLIP_LP+LNVision Backbone=ViT-B/162022.11 | 97.94 | — | — | — | 92.08 | — | — | — | |
| PEMnE-BMS*Vision Backbone=DenseNet1212022.11 | 91.53 | — | — | — | 85.54 | — | — | — | |
| BECLRBackbone=RN50, Setting=Unsupervised2024.02 | 87.82 | — | — | — | 80.57 | — | — | — | |
| Meta-DM + UniSiam + distBackbone=RN50, Setting=Unsupervised, Extra Synthetic Data=true2024.02 | 85.29 | — | — | — | 66.68 | — | — | — | |
| BECLRBackbone=RN18, Setting=Unsupervised2024.02 | 84.93 | — | — | — | 75.74 | — | — | — | |
| H-OTBackbone=WRN28+RTloss2022.10 | 84.36 | — | — | — | 69.04 | — | — | — | |
| METAQDABackbone=WRN2021.01 | 84.28 | — | — | — | 67.83 | — | — | — | |
| Meta-DM + UniSiam + distBackbone=RN18, Setting=Unsupervised, Extra Synthetic Data=true2024.02 | 83.97 | — | — | — | 65.64 | — | — | — | |
| NNCLRBackbone=RN50, Setting=Unsupervised2024.02 | 83.31 | — | — | — | 65.42 | — | — | — | |
| UniSiam + distBackbone=RN50, Setting=Unsupervised2024.02 | 83.22 | — | — | — | 65.33 | — | — | — | |
| S2M2Backbone=WRN, Note=Fixed feature method2021.01 | 83.18 | — | — | — | 64.93 | — | — | — | |
| PDA-NetBackbone=RN50, Setting=Unsupervised2024.02 | 83.11 | — | — | — | 63.84 | — | — | — | |
| Free-LunchBackbone=WRN28+RTloss2022.10 | 82.88 | — | — | — | 68.57 | — | — | — | |
| H-OTBackbone=ResNet122022.10 | 82.87 | — | — | — | 65.63 | — | — | — | |
| SwAVBackbone=RN50, Setting=Unsupervised2024.02 | 82.76 | — | — | — | 63.34 | — | — | — | |
| DeepEMDBackbone=ResNet122022.10 | 82.41 | — | — | — | 65.91 | — | — | — | |
| UniSiam + distBackbone=RN18, Setting=Unsupervised2024.02 | 82.26 | — | — | — | 64.1 | — | — | — | |
| Negative-CosineBackbone=ResNet122022.10 | 81.57 | — | — | — | 63.85 | — | — | — | |
| Free-LunchBackbone=ResNet12, Implementation=H-OT paper reproduction2022.10 | 81.15 | — | — | — | 64.73 | — | — | — | |
| E3BMBackbone=ResNet252022.10 | 81 | — | — | — | 64.3 | — | — | — | |
| METAQDABackbone=ResNet-182021.01 | 80.98 | — | — | — | 65.12 | — | — | — | |
| NNCLRBackbone=RN18, Setting=Unsupervised, reproduction=true2024.02 | 80.75 | — | — | — | 63.33 | — | — | — | |
| S2M2Backbone=ResNet-18, Note=Fixed feature method2021.01 | 80.58 | — | — | — | 64.06 | — | — | — | |
| SIMPLESHOTBackbone=WRN, Note=Fixed feature method2021.01 | 80.33 | — | — | — | 63.5 | — | — | — | |
| SIMPLESHOTBackbone=ResNet-18, Note=Fixed feature method2021.01 | 80.02 | — | — | — | 62.85 | — | — | — | |
| MetaOptNetBackbone=RN18, Setting=Supervised2024.02 | 80 | — | — | — | 64.09 | — | — | — | |
| METAOPTBackbone=ResNet-12 (Wide CNN), Note=Requires gradient-based optimisation at meta-test time2021.01 | 80 | — | — | — | 64.09 | — | — | — | |
| SimSiamBackbone=RN18, Setting=Unsupervised2024.02 | 79.85 | — | — | — | 62.8 | — | — | — | |
| SimCLRBackbone=RN18, Setting=Unsupervised2024.02 | 79.66 | — | — | — | 62.58 | — | — | — | |
| SURBackbone=ResNet-122021.01 | 79.25 | — | — | — | 60.79 | — | — | — | |
| Laplacian EigenmapsBackbone=RN18, Setting=Unsupervised2024.02 | 78.79 | — | — | — | 59.47 | — | — | — | |
| SwAVBackbone=RN18, Setting=Unsupervised2024.02 | 78.23 | — | — | — | 59.84 | — | — | — | |
| AFHNBackbone=ResNet-182021.01 | 78.16 | — | — | — | 62.38 | — | — | — | |
| R2D2Backbone=ResNet-122021.01 | 78.15 | — | — | — | 59.38 | — | — | — | |
| ProtoNetBackbone=ResNet122022.10 | 78.02 | — | — | — | 60.37 | — | — | — | |
| LEOBackbone=WRN282022.10 | 77.59 | — | — | — | 61.76 | — | — | — | |
| LEOBackbone=WRN, Note=Requires gradient-based optimisation at meta-test time2021.01 | 77.59 | — | — | — | 61.78 | — | — | — | |
| DCEMBackbone=ResNet-182021.01 | 77.28 | — | — | — | 58.71 | — | — | — | |
| RELATIONNET2Backbone=ResNet-122021.01 | 77.15 | — | — | — | 63.92 | — | — | — | |
| UNRAVELLINGBackbone=ResNet-12 (Wide CNN)2021.01 | 77.05 | — | — | — | 59.37 | — | — | — | |
| TADAMBackbone=ResNet-122021.01 | 76.7 | — | — | — | 58.5 | — | — | — | |
| TAP NETBackbone=ResNet-122021.01 | 76.36 | — | — | — | 61.65 | — | — | — | |
| MatchNetBackbone=ResNet122022.10 | 75.99 | — | — | — | 63.08 | — | — | — | |
| HMSBackbone=RN18, Setting=Unsupervised2024.02 | 75.77 | — | — | — | 58.2 | — | — | — | |
| Baseline++Backbone=ResNet182022.10 | 75.68 | — | — | — | 51.87 | — | — | — | |
| BASELINE++Backbone=ResNet-182021.01 | 75.68 | — | — | — | 51.87 | — | — | — | |
| MTLBackbone=ResNet-12 (Wide CNN)2021.01 | 75.5 | — | — | — | 61.2 | — | — | — | |
| UBC-FSLBackbone=RN50, Setting=Unsupervised2024.02 | 75.4 | — | — | — | 56.2 | — | — | — | |
| AM3Backbone=ResNet-122021.01 | 75.2 | — | — | — | 65.21 | — | — | — | |
| ProtoNet + POTBackbone=ResNet342022.10 | 75.15 | — | — | — | — | — | — | — | |
| ProtoNetBackbone=RN34, Setting=Supervised2024.02 | 74.65 | — | — | — | 53.9 | — | — | — | |
| ProtoNetBackbone=ResNet342022.10 | 73.99 | — | — | — | — | — | — | — | |
| TrainProtoBackbone=RN50, Setting=Unsupervised2024.02 | 73.94 | — | — | — | 58.92 | — | — | — | |
| PPABackbone=WRN2021.01 | 73.74 | — | — | — | 59.6 | — | — | — | |
| PROTONETBackbone=ResNet-182021.01 | 73.68 | — | — | — | 54.16 | — | — | — | |
| Transductive CNAPSBackbone=RN18, Setting=Supervised2024.02 | 73.1 | — | — | — | 55.6 | — | — | — | |
| DYNAMICFSLBackbone=Conv-4, Optimization=two-step optimization with attention2021.01 | 72.81 | — | — | — | 56.2 | — | — | — | |
| METAQDABackbone=Conv-42021.01 | 72.64 | — | — | — | 56.41 | — | — | — | |
| SAMPTransferBackbone=Conv4b, Setting=Unsupervised2024.02 | 72.52 | — | — | — | 61.02 | — | — | — | |
| CAMLBackbone=ResNet-122021.01 | 72.35 | — | — | — | 59.23 | — | — | — | |
| GCRBackbone=Conv-42021.01 | 72.34 | — | — | — | 53.21 | — | — | — | |
| DYNAMIC FSLBackbone=ResNet-122021.01 | 70.13 | — | — | — | 55.45 | — | — | — | |
| RELATIONNETBackbone=ResNet-182021.01 | 69.83 | — | — | — | 52.48 | — | — | — | |
| SNAILBackbone=ResNet-122021.01 | 68.88 | — | — | — | 55.71 | — | — | — | |
| R2D2Backbone=Conv-4 (Wide CNN)2021.01 | 68.7 | — | — | — | 51.9 | — | — | — | |
| MatchNet + POTBackbone=ResNet342022.10 | 68.51 | — | — | — | — | — | — | — | |
| MatchNetBackbone=ResNet342022.10 | 68.32 | — | — | — | — | — | — | — | |
| PROTONETBackbone=Conv-42021.01 | 68.2 | — | — | — | 49.42 | — | — | — | |
| RELATIONNET2Backbone=Conv-42021.01 | 67.63 | — | — | — | 53.48 | — | — | — | |
| VERSABackbone=Conv-42021.01 | 67.37 | — | — | — | 53.4 | — | — | — | |
| CPNWCPBackbone=RN18, Setting=Unsupervised2024.02 | 67.36 | — | — | — | 53.14 | — | — | — | |
| SIMPLESHOTBackbone=Conv-4, Note=Fixed feature method2021.01 | 66.92 | — | — | — | 49.69 | — | — | — | |
| TPNBackbone=Conv-42021.01 | 66.59 | — | — | — | 52.78 | — | — | — | |
| LF2CSBackbone=RN12, Setting=Unsupervised2024.02 | 66.44 | — | — | — | 47.93 | — | — | — | |
| BASELINE++Backbone=Conv-42021.01 | 66.43 | — | — | — | 48.24 | — | — | — | |
| GNNBackbone=Conv-42021.01 | 66.41 | — | — | — | 50.33 | — | — | — | |
| MAMLBackbone=RN34, Setting=Supervised2024.02 | 65.9 | — | — | — | 51.46 | — | — | — | |
| CAVIABackbone=Conv-42021.01 | 65.85 | — | — | — | 51.82 | — | — | — | |
| MAMLBackbone=ResNet182022.10 | 65.72 | — | — | — | 49.61 | — | — | — | |
| PsCotBackbone=RN18, Setting=Unsupervised2024.02 | 65.48 | — | — | — | 47.24 | — | — | — | |
| RELATIONNETBackbone=Conv-42021.01 | 65.32 | — | — | — | 50.44 | — | — | — | |
| C3LRBackbone=Conv4, Setting=Unsupervised2024.02 | 64.81 | — | — | — | 47.92 | — | — | — | |
| METASSLBackbone=Conv-42021.01 | 64.39 | — | — | — | 50.41 | — | — | — | |
| METASGDBackbone=Conv-4, Note=Requires gradient-based optimisation at meta-test time2021.01 | 64.03 | — | — | — | 50.47 | — | — | — | |
| MAMLBackbone=Conv-4, Note=Requires gradient-based optimisation at meta-test time2021.01 | 63.11 | — | — | — | 48.7 | — | — | — | |
| ProtoTransferBackbone=Conv4, Setting=Unsupervised2024.02 | 62.99 | — | — | — | 45.67 | — | — | — | |
| METALSTMBackbone=Conv-42021.01 | 60.6 | — | — | — | 43.44 | — | — | — | |
| Meta-GMVAEBackbone=Conv4, Setting=Unsupervised2024.02 | 55.73 | — | — | — | 42.82 | — | — | — | |
| FREE5Data recovery process=Inversion, Adaptation steps=5, Time (GPU hours)=0.41h2024.05 | 45.45 | — | — | — | 33.03 | — | — | — | |
| BiDf-MKDData recovery process=Inversion, Time (GPU hours)=8.87h2024.05 | 42.3 | — | — | — | 30.66 | — | — | — | |
| FREE2Data recovery process=Inversion, Adaptation steps=2, Time (GPU hours)=0.25h2024.05 | 41.6 | — | — | — | 30.06 | — | — | — | |
| PURERData recovery process=Inversion, Time (GPU hours)=1.31h2024.05 | 40.86 | — | — | — | 31.14 | — | — | — | |
| DROData recovery process=Non-inversion2024.05 | 30.19 | — | — | — | 27.56 | — | — | — | |
| RandomData recovery process=Non-inversion2024.05 | 28.1 | — | — | — | 25.06 | — | — | — | |
| AverageData recovery process=Non-inversion2024.05 | 27.49 | — | — | — | 23.79 | — | — | — | |
| OTAData recovery process=Non-inversion2024.05 | 27.22 | — | — | — | 24.22 | — | — | — | |
| Best-ModelData recovery process=Non-inversion2024.05 | 26.26 | — | — | — | 22.86 | — | — | — | |
| ABHFA-NetN-way=5, K-shot=1, Backbone=ResNet122025.10 | — | — | — | — | — | — | — | 73.4 | |
| ABHFA-NetN-way=5, K-shot=5, Backbone=ResNet122025.10 | — | — | — | — | — | — | — | 87.4 |