5-way Few-Shot Classification Accuracy on miniImageNet (test)
81.94AccuracyRobust 20 Full
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Robust 20 FullInput size=80, Network=WideResNet2019.03 | 81.94 | — | — | — | |
| Robust 20 FullInput size=224, Network=ResNet2019.03 | 81.59 | — | — | — | |
| Robust 20-dist++Input size=224, Network=ResNet2019.03 | 81.19 | — | — | — | |
| Robust 20-dist++Input size=80, Network=WideResNet2019.03 | 81.17 | — | — | — | |
| CTMBackbone=ResNet-18, Shot=5-shot2020.04 | 80.5 | — | — | — | |
| SimpleShot(UN)Number of shots=52020.11 | 80.43 | — | — | — | |
| Gidaris et al.Backbone=WRN-28-10, Shot=5-shot2020.04 | 79.9 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=5-shot, K=5, learning_rate=1e-32020.04 | 79.2 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=5-shot, K=3, learning_rate=1e-32020.04 | 78.9 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=5-shot, K=1, learning_rate=1e-32020.04 | 78.8 | — | — | — | |
| MetaOptNet-SVMBackbone=ResNet-12, Shot=5-shot2020.04 | 78.6 | — | — | — | |
| FEATInput size=80, Network=WideResNet2019.03 | 78.32 | — | — | — | |
| MetaOptNet-RRBackbone=ResNet-12, Shot=5-shot2020.04 | 77.9 | — | — | — | |
| LEOBackbone=WRN-28-10, Shot=5-shot2020.04 | 77.6 | — | — | — | |
| LEOInput size=80, Network=WideResNet2019.03 | 77.59 | — | — | — | |
| LEONumber of shots=52020.11 | 77.59 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=5-shot, K=0, Feature=Pre-trained2020.04 | 77.5 | — | — | — | |
| Robust 20 FullInput size=84, Network=ResNet2019.03 | 76.9 | — | — | — | |
| TADAMFeature extractor=ResNet-12 (pre), Additional layers=72 fully connected layers2018.12 | 76.7 | — | — | — | |
| TADAMInput size=84, Network=ResNet2019.03 | 76.7 | — | — | — | |
| TADAMBackbone=ResNet-12, Shot=5-shot2020.04 | 76.7 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=5-shot, K=5, learning_rate=1e-32020.04 | 75.73 | — | — | — | |
| Cosine ClassifierInput size=224, Network=ResNet2019.03 | 75.68 | — | — | — | |
| Robust 20-dist++Input size=84, Network=ResNet2019.03 | 75.62 | — | — | — | |
| Supervised MAMLEmbedding=N/A, K_tr=50, K_val=152020.06 | 75.54 | — | — | — | |
| MTLFeature extractor=ResNet-12 (pre), Meta-batch type=HT meta-batch, Protocol=SS [Θ; θ]2018.12 | 75.5 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=5-shot, K=3, learning_rate=1e-32020.04 | 75.43 | — | — | — | |
| MTLFeature extractor=ResNet-12 (pre), Meta-batch type=meta-batch, Protocol=SS [Θ; θ]2018.12 | 74.3 | — | — | — | |
| Linear ClassifierInput size=224, Network=ResNet2019.03 | 74.27 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=5-shot, K=1, learning_rate=1e-32020.04 | 74.12 | — | — | — | |
| Ours-WRNNumber of shots=5, Backbone=WRN-28-102017.06 | 73.74 | — | — | — | |
| PPAInput size=80, Network=WideResNet2019.03 | 73.74 | — | — | — | |
| QiaoNumber of shots=52020.11 | 73.74 | — | — | — | |
| Qiao et al.Backbone=WRN-28-10, Shot=5-shot2020.04 | 73.7 | — | — | — | |
| Delta-encoderFeature extractor=VGG-16 (pre)2018.12 | 73.6 | — | — | — | |
| MAML deep, HTFeature extractor=ResNet-12 (pre), Meta-batch type=HT meta-batch, Protocol=FT [Θ; θ]2018.12 | 73.1 | — | — | — | |
| Cosine + AttentionInput size=224, Network=ResNet2019.03 | 73 | — | — | — | |
| Supervised ProtoNetsEmbedding=N/A, K_tr=50, K_val=152020.06 | 72.04 | — | — | — | |
| adaResNetFeature extractor=ResNet-12, Additional layers=1 convolutional layer2018.12 | 71.94 | — | — | — | |
| adaCNNNumber of shots=52020.11 | 71.94 | — | — | — | |
| MTUNetBackbone=WRN, Number of shots=52020.11 | 71.93 | — | — | — | |
| Gidaris et al.Backbone=Conv-4-64, Shot=5-shot2020.04 | 71.9 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=5-shot, K=0, learning_rate=1e-32020.04 | 71.48 | — | — | — | |
| Supervised MAMLEmbedding=N/A, K_tr=20, K_val=152020.06 | 71.03 | — | — | — | |
| SIBBackbone=Conv-4-64, Shot=5-shot, K=3, learning_rate=1e-32020.04 | 70.7 | — | — | — | |
| MTUNetBackbone=ResNet-18, Number of shots=52020.11 | 70.22 | — | — | — | |
| GidarisNumber of shots=52020.11 | 70.13 | — | — | — | |
| Supervised ProtoNetsEmbedding=N/A, K_tr=20, K_val=152020.06 | 70.05 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=1-shot, K=5, learning_rate=1e-32020.04 | 70 | — | — | — | |
| TPNBackbone=Conv-4-64, Shot=5-shot2020.04 | 69.9 | — | — | — | |
| RelationNetNumber of shots=5, variant=22020.11 | 69.83 | — | — | — | |
| CACTUS MAMLEmbedding=DeepCluster, K_tr=50, K_val=152020.06 | 69.64 | — | — | — | |
| Adv. ResNetFeature extractor=WRN-40 (pre)2018.12 | 69.6 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=1-shot, K=3, learning_rate=1e-32020.04 | 69.6 | — | — | — | |
| LASIUM-N-GAN-MAMLEmbedding=N/A, K_tr=50, K_val=152020.06 | 69.13 | — | — | — | |
| SNAILBackbone=ResNet-12, Shot=5-shot2020.04 | 68.9 | — | — | — | |
| SNAILFeature extractor=ResNet-12 (pre)2018.12 | 68.88 | — | — | — | |
| SNAILNumber of shots=52020.11 | 68.88 | — | — | — | |
| MetaGANFeature extractor=ResNet-122018.12 | 68.63 | — | — | — | |
| ProtoNetsFeature extractor=4 CONV2018.12 | 68.2 | — | — | — | |
| Prototypical NetBackbone=Conv-4-64, Shot=5-shot2020.04 | 68.2 | — | — | — | |
| ProtoNetNumber of shots=52020.11 | 68.2 | — | — | — | |
| R2-D2Number of shots=52020.11 | 68.2 | — | — | — | |
| Ours-SimpleNumber of shots=5, Backbone=Simple Convolutional Network2017.06 | 67.87 | — | — | — | |
| SIBBackbone=WRN-28-10, Shot=1-shot, K=1, learning_rate=1e-32020.04 | 67.3 | — | — | — | |
| UMTRA MAMLEmbedding=N/A, K_tr=50, K_val=152020.06 | 67.15 | — | — | — | |
| SIBBackbone=Conv-4-64, Shot=5-shot, K=0, Feature=Pre-trained2020.04 | 67 | — | — | — | |
| CACTUS MAMLEmbedding=BIGAN, K_tr=50, K_val=152020.06 | 66.91 | — | — | — | |
| GNNNumber of shots=52020.11 | 66.41 | — | — | — | |
| GNNBackbone=Conv-4-64, Shot=5-shot2020.04 | 66.4 | — | — | — | |
| R2-D2Backbone=Conv-4-64, Shot=5-shot2020.04 | 65.4 | — | — | — | |
| CompareNetsFeature extractor=4 CONV2018.12 | 65.32 | — | — | — | |
| RelationNetNumber of shots=5, variant=12020.11 | 65.32 | — | — | — | |
| Relation NetBackbone=Conv-4-64, Shot=5-shot2020.04 | 65.3 | — | — | — | |
| Linear ClassifierEmbedding=DeepCluster, K_tr=50, K_val=152020.06 | 65.28 | — | — | — | |
| LASIUM-N-GAN-MAMLEmbedding=N/A, K_tr=20, K_val=152020.06 | 65.17 | — | — | — | |
| Bilevel ProgrammingFeature extractor=ResNet-12, Additional layers=2 convolutional layers2018.12 | 64.53 | — | — | — | |
| MAML, HTFeature extractor=4 CONV, Meta-batch type=HT meta-batch, Protocol=FT [Θ; θ]2018.12 | 64.1 | — | — | — | |
| CTMBackbone=ResNet-18, Shot=1-shot2020.04 | 64.1 | — | — | — | |
| Meta SGDNumber of shots=52020.11 | 64.03 | — | — | — | |
| K-nearest neighborsEmbedding=DeepCluster, K_tr=50, K_val=152020.06 | 63.9 | — | — | — | |
| CACTUS MAMLEmbedding=DeepCluster, K_tr=20, K_val=152020.06 | 63.84 | — | — | — | |
| CACTUS ProtoNetsEmbedding=DeepCluster, K_tr=50, K_val=152020.06 | 63.55 | — | — | — | |
| CACTUS ProtoNetsEmbedding=BIGAN, K_tr=50, K_val=152020.06 | 63.27 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=1-shot, K=5, learning_rate=1e-32020.04 | 63.26 | — | — | — | |
| MAMLNumber of shots=52017.06 | 63.11 | — | — | — | |
| MAMLFeature extractor=4 CONV2018.12 | 63.11 | — | — | — | |
| MAMLNumber of shots=52020.11 | 63.11 | — | — | — | |
| MAMLBackbone=Conv-4-64, Shot=5-shot2020.04 | 63.1 | — | — | — | |
| Gidaris et al.Backbone=WRN-28-10, Shot=1-shot2020.04 | 62.9 | — | — | — | |
| MetaOptNet-SVMBackbone=ResNet-12, Shot=1-shot2020.04 | 62.6 | — | — | — | |
| SIBBackbone=Conv-4-128, Shot=1-shot, K=3, learning_rate=1e-32020.04 | 62.59 | — | — | — | |
| Supervised ProtoNetsEmbedding=N/A, K_tr=5, K_val=152020.06 | 62.29 | — | — | — | |
| ProtoNetsClustering=N/A, Training Protocol=Supervised (Upper Bound), Backbone=4-layer CNN2020.04 | 62.29 | — | — | — | |
| Supervised MAMLEmbedding=N/A, K_tr=5, K_val=152020.06 | 62.13 | — | — | — | |
| MAMLClustering=N/A, Training Protocol=Supervised (Upper Bound), Backbone=4-layer CNN2020.04 | 62.13 | — | — | — | |
| LEOBackbone=WRN-28-10, Shot=1-shot2020.04 | 61.8 | — | — | — | |
| LEONumber of shots=12020.11 | 61.76 | — | — | — | |
| CACTUS ProtoNetsEmbedding=DeepCluster, K_tr=20, K_val=152020.06 | 61.54 | — | — | — | |
| LASIUM-N-GAN-ProtoNetsEmbedding=N/A, K_tr=50, K_val=152020.06 | 61.43 | — | — | — |