Few-shot classification on miniImageNet standard (test)
98.345-way 1-shot AccEfficientFSL
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| EfficientFSLBackbone=ViT-B, Params=2.48M, Pre-training=ImageNet21K2026.01 | 98.34 | 99.12 | — | |
| EfficientFSLBackbone=ViT-B, Params=2.48M, Pre-training=ImageNet1K2026.01 | 97.57 | 98.96 | — | |
| EfficientFSLBackbone=ViT-S, Params=1.25M, Pre-training=ImageNet1K2026.01 | 97.4 | 99.05 | — | |
| Full Fine-TuningBackbone=ViT-B, Params=85.8M2026.01 | 90.51 | 95.01 | — | |
| Full Fine-TuningBackbone=ViT-S, Params=21.7M2026.01 | 89.41 | 95.59 | — | |
| FewVSBackbone=ViT-S, Params=21.7M2026.01 | 86.8 | 90.32 | — | |
| MetaF.Backbone=ViT-S, Params=24.5M2026.01 | 84.78 | 91.39 | — | |
| SemFewBackbone=Swin-T, Params=88.0M2026.01 | 78.94 | 86.49 | — | |
| Ours_Image900_SSLEmbedding Net=AmdimNet, Meta-learning=true2019.11 | 76.82 | 90.98 | — | |
| KTPPBackbone=Visformer-T, Params=11.1M2026.01 | 76.71 | 86.46 | — | |
| SCAM-NetBackbone=ViT-S, Params=21.7M2026.01 | 75.93 | 89.75 | — | |
| SPBackbone=Visformer-T, Params=10.3M2026.01 | 72.31 | 83.42 | — | |
| EPNetBackbone=WRN-28-10, Params=37582K2020.03 | 70.74 | 84.34 | — | |
| ProtoNetBackbone=ResNet-12, Protocol=S/T, Selection Ratio=10%2021.04 | 68.03 | 82.53 | — | |
| FewTUREBackbone=ViT-S, Params=21.7M2026.01 | 68.02 | 84.51 | — | |
| DPGNBackbone=ResNet12, Testing Mode=Transductive2020.03 | 67.77 | 84.6 | — | |
| ProtoNetBackbone=ResNet-12, Protocol=S/T, Selection Ratio=5%2021.04 | 67.35 | 81.67 | — | |
| DPGNBackbone=WRN, Testing Mode=Transductive2020.03 | 67.24 | 83.72 | — | |
| CANBackbone=RESNET-12, Params=8026K2020.03 | 67.19 | 80.64 | — | |
| AM3 (Prototypical Network) + TRAMLBackbone=ResNet12, Type=Metric2020.05 | 67.1 | 79.54 | — | |
| FEATBackbone=ResNet-122021.04 | 66.78 | 82.05 | — | |
| DPGNBackbone=ResNet18, Testing Mode=Transductive2020.03 | 66.63 | 84.07 | — | |
| EPNetBackbone=RESNET-12, Params=7989K2020.03 | 66.5 | 81.06 | — | |
| FRNBackbone=ResNet-122021.04 | 66.45 | 82.83 | — | |
| DPGNBackbone=ConvNet, Testing Mode=Transductive2020.03 | 66.01 | 82.83 | — | |
| --NetBackbone=WRN-28-10, Params=37582K2020.03 | 65.98 | 82.22 | — | |
| DeepEMDBackbone=ResNet-122021.04 | 65.91 | 82.41 | — | |
| --NetBackbone=RESNET-12, Params=7989K2020.03 | 65.66 | 81.28 | — | |
| ProtoNetBackbone=ResNet-12, Protocol=S/Q, Status=re-implemented2021.04 | 65.3 | 79.93 | — | |
| AM3 (Prototypical Network)Backbone=ResNet12, Type=Metric2020.05 | 65.21 | 75.2 | — | |
| Manifold mixupBackbone=WRN-28-10, Params=37582K2020.03 | 64.93 | 83.18 | — | |
| MetaFun-DFPFeature Extraction=Deep residual networks, Data Augmentation=Yes2019.12 | 64.13 | 80.82 | — | |
| MetaOptNet-SVMFeature Extraction=Deep residual networks, Data Augmentation=Yes2019.12 | 64.09 | 80 | — | |
| Ours_Mini80_SSLEmbedding Net=AmdimNet, Meta-learning=true2019.11 | 64.03 | 81.15 | — | |
| LEOFeature Extraction=Deep residual networks, Data Augmentation=Yes2019.12 | 63.97 | 79.49 | — | |
| Qiao et al.Feature Extraction=Deep residual networks, Data Augmentation=Yes2019.12 | 63.62 | 78.83 | — | |
| MetaFun-KFPFeature Extraction=Deep residual networks, Data Augmentation=Yes2019.12 | 63.39 | 80.81 | — | |
| FEATBackbone=ResNet12, Testing Mode=Inductive2020.03 | 62.96 | 78.49 | — | |
| WDAE-GNNBackbone=WRN-28-10, Params=48855K2020.03 | 62.96 | 78.85 | — | |
| CC+rotBackbone=WRN-28-10, Params=37582K2020.03 | 62.93 | 79.87 | — | |
| Robust-20++Backbone=WRN-28-10, Params=37582K2020.03 | 62.8 | 80.85 | — | |
| MetaOptNetBackbone=ResNet12, Testing Mode=Inductive2020.03 | 62.64 | 78.63 | — | |
| MetaOptNet-SVMBackbone=ResNet12, Type=Gradient2020.05 | 62.64 | 78.63 | — | |
| MetaOpt-SVMBackbone=RESNET-12, Params=12415K2020.03 | 62.64 | 78.6 | — | |
| DenseBackbone=ResNet12, Testing Mode=Inductive2020.03 | 62.53 | 78.95 | — | |
| DCBackbone=ResNet12, Type=Metric2020.05 | 62.53 | 78.95 | — | |
| MetaFun-DFPFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 62.12 | 77.78 | — | |
| CTMBackbone=ResNet18, Testing Mode=Inductive2020.03 | 62.05 | 78.63 | — | |
| LEOBackbone=WRN, Testing Mode=Inductive2020.03 | 61.76 | 77.59 | — | |
| LEOBackbone=WRN-28-10, Params=37582K2020.03 | 61.76 | 77.59 | — | |
| LEOFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 61.76 | 77.59 | — | |
| LEOEmbedding Net=Wide-ResNet28, Meta-learning=true2019.11 | 61.76 | 77.59 | — | |
| TapNetBackbone=ResNet12, Testing Mode=Inductive2020.03 | 61.65 | 76.36 | — | |
| TapNetBackbone=ResNet12, Type=Metric2020.05 | 61.65 | 76.36 | — | |
| Meta-TransferBackbone=ResNet12, Testing Mode=Inductive2020.03 | 61.2 | 75.53 | — | |
| MTLBackbone=ResNet12, Type=Gradient2020.05 | 61.2 | 75.5 | — | |
| MTLBackbone=RESNET-12, Params=8286K2020.03 | 61.2 | 75.5 | — | |
| MTLEmbedding Net=ResNet12, Meta-learning=true2019.11 | 61.2 | 75.5 | — | |
| MetaFun-KFPFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 61.16 | 78.2 | — | |
| WDAEBackbone=WRN, Testing Mode=Inductive2020.03 | 61.07 | 76.75 | — | |
| ProtoNetBackbone=ResNet-12, Protocol=S/Q2021.04 | 60.37 | 78.02 | — | |
| Prototypical Network + TRAMLBackbone=ResNet12, Type=Metric2020.05 | 60.31 | 77.94 | — | |
| MAMLBackbone=ResNet-12, Protocol=S/T, Selection Ratio=10%2021.04 | 60.06 | 76.34 | — | |
| Edge-labelBackbone=ConvNet, Testing Mode=Transductive2020.03 | 59.63 | 76.34 | — | |
| Qiao et al.Feature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 59.6 | 73.74 | — | |
| Qiao-WRNEmbedding Net=Wide-ResNet28, Meta-learning=true2019.11 | 59.6 | 73.74 | — | |
| TPNBackbone=RESNET-12, Params=8284K2020.03 | 59.46 | 75.65 | — | |
| EPNetBackbone=CONV-4, Params=112K2020.03 | 59.32 | 72.95 | — | |
| CAMLBackbone=ResNet12, Type=Gradient2020.05 | 59.23 | 72.35 | — | |
| MAMLBackbone=ResNet-12, Protocol=S/T, Selection Ratio=5%2021.04 | 59.14 | 75.77 | — | |
| Shot-FreeBackbone=ResNet12, Testing Mode=Inductive2020.03 | 59.04 | 77.64 | — | |
| ECMSFMTBackbone=ResNet12, Type=Metric2020.05 | 59 | 77.46 | — | |
| MAMLBackbone=ResNet-12, Protocol=S/Q, Status=re-implemented2021.04 | 58.84 | 74.62 | — | |
| Self-Jig(SVM)Embedding Net=ResNet50, Meta-learning=true2019.11 | 58.8 | 76.71 | — | |
| TADAMBackbone=ResNet12, Testing Mode=Inductive2020.03 | 58.5 | 76.7 | — | |
| TADAMBackbone=ResNet12, Type=Metric2020.05 | 58.5 | 76.7 | — | |
| TADAMBackbone=RESNET-12, Params=7989K2020.03 | 58.5 | 76.7 | — | |
| TADAMFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 58.5 | 76.7 | — | |
| TADAMEmbedding Net=ResNet12, Meta-learning=true2019.11 | 58.5 | 76.7 | — | |
| Robust-20++Backbone=RESNET-12, Params=11174K2020.03 | 58.11 | 75.24 | — | |
| --NetBackbone=CONV-4, Params=112K2020.03 | 57.18 | 72.57 | — | |
| Munkhdalai et al.Feature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 57.1 | 70.04 | — | |
| Prototypical NetworkBackbone=ResNet12, Type=Metric2020.05 | 56.52 | 74.28 | — | |
| ProtoNets++Backbone=RESNET-12, Params=7989K2020.03 | 56.52 | 74.28 | — | |
| ProtoNet+Embedding Net=ResNet12, Meta-learning=true2019.11 | 56.5 | 74.2 | — | |
| Bauer et al.Feature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 56.3 | 73.9 | — | |
| Dis. k-shotEmbedding Net=ResNet34, Meta-learning=true2019.11 | 56.3 | 73.9 | — | |
| DynamicBackbone=ConvNet, Testing Mode=Inductive2020.03 | 56.2 | 71.94 | — | |
| Dynamic FSLBackbone=4Conv, Type=Metric2020.05 | 56.2 | 73 | — | |
| SNAILBackbone=ResNet12, Testing Mode=Inductive2020.03 | 55.71 | 68.88 | — | |
| SNAILFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 55.71 | 68.88 | — | |
| SNAILEmbedding Net=ResNet12, Meta-learning=true2019.11 | 55.71 | 68.88 | — | |
| TPNBackbone=ConvNet, Testing Mode=Transductive2020.03 | 55.51 | 69.86 | — | |
| CC+rotBackbone=CONV-4, Params=112K2020.03 | 54.83 | 71.86 | — | |
| LCCBackbone=4Conv, Type=Gradient2020.05 | 54.6 | 71.1 | — | |
| DN4Embedding Net=ResNet12, Meta-learning=true2019.11 | 54.37 | 74.44 | — | |
| Meta-SGDFeature Extraction=Deep residual networks, Data Augmentation=No2019.12 | 54.24 | 70.86 | — | |
| FEATEmbedding Net=ResNet50, Meta-learning=true2019.11 | 53.8 | 76 | — | |
| TPNBackbone=CONV-4, Params=171K2020.03 | 53.75 | 69.43 | — | |
| Memory Matching NetworkBackbone=4Conv, Type=Metric2020.05 | 53.37 | 66.97 | — |