Image Classification on SVHN 1000 labels (test)
1.89Error RateUDA
Evaluation Results
| Method | Links | |
|---|---|---|
| UDA2021.10 | 1.89 | |
| FixMatch2021.10 | 1.96 | |
| UDA (RandAugment)Model=WRN-28-2, # Param=1.5M2019.04 | 2.23 | |
| 100% SupervisedBackbone=ConvLarge, labeled_data=100%2019.08 | 2.88 | |
| MixMatchModel=WRN-28-2, # Param=1.5M2019.04 | 2.89 | |
| Mean Teacher2021.10 | 3.27 | |
| DCT with 8 ViewsBackbone=ConvLarge2019.08 | 3.29 | |
| Triple-GAN-V2Classifier=13-layer CNN [32], Data Augmentation=Standard2019.12 | 3.45 | |
| ICTModel=WRN-28-2, # Param=1.5M2019.04 | 3.53 | |
| VAdD (KL) + VATBackbone=ConvLarge2019.08 | 3.55 | |
| R2-D2Backbone=ConvLarge2019.08 | 3.64 | |
| MixMatch2021.10 | 3.69 | |
| VAT+EntMinBackbone=ConvLarge2019.08 | 3.86 | |
| VAT+EntClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.12 | 3.86 | |
| SNTGClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.12 | 3.86 | |
| VAT + EntMinModel=Conv-Large, # Param=3.1M2019.04 | 3.86 | |
| SNTGModel=Conv-Large, # Param=3.1M2019.04 | 3.86 | |
| ICTModel=Conv-Large, # Param=3.1M2019.04 | 3.89 | |
| Mean TeacherBackbone=ConvLarge2019.08 | 3.95 | |
| MTClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.12 | 3.95 | |
| Mean TeacherModel=Conv-Large, # Param=3.1M2019.04 | 3.95 | |
| VAT2021.10 | 4.11 | |
| VAdD (KL)Backbone=ConvLarge2019.08 | 4.16 | |
| Temporal EnsemblingBackbone=ConvLarge2019.08 | 4.42 | |
| MT (our code base)Classifier=13-layer CNN [32], Data Augmentation=Standard2019.12 | 4.48 | |
| Π-ModelModel=Conv-Large, # Param=3.1M2019.04 | 4.82 | |
| II modelClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.12 | 4.95 | |
| ReMixMatch2021.10 | 5.16 | |
| VATClassifier=13-layer CNN (similar), Data Augmentation=Standard2019.12 | 5.42 | |
| Shake-Shake (26 2x96d) + AutoAugmentBackbone=Shake-Shake (26 2x96d), Augmentation=AutoAugment2018.05 | 5.9 | |
| LGA + VATModel=WRN-28-2, # Param=1.5M2019.04 | 6.58 | |
| FlexMatch2021.10 | 6.72 | |
| Π-Model2021.10 | 7.16 | |
| Pseudo-LabelModel=WRN-28-2, # Param=1.5M2019.04 | 7.62 | |
| Wide-ResNet-28-10 + AutoAugmentBackbone=Wide-ResNet-28-10, Augmentation=AutoAugment2018.05 | 8.2 | |
| Pseudo-Labeling2021.10 | 9.4 | |
| Triple-GAN-V2Data augmentation=true, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 10.17 | |
| Triple-GAN-V2Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 11.08 | |
| MTData augmentation=true, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 11.08 | |
| Only 1000 labeled imagesBackbone=ConvLarge, labeled_data=1000 samples2019.08 | 11.27 | |
| MTData augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 11.68 | |
| Shake-Shake (26 2x96d)Backbone=Shake-Shake (26 2x96d), Augmentation=Baseline2018.05 | 12.3 | |
| Wide-ResNet-28-10Backbone=Wide-ResNet-28-10, Augmentation=Baseline2018.05 | 13.2 | |
| Triple-GAN-V1Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 21.44 | |
| DCGANclassifier=L2-SVM2015.11 | 22.48 | |
| CNN (our code base)Data augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 22.72 | |
| SWWAEdropout=true2015.11 | 23.56 | |
| SWWAEN=1000, Dropout=true, lambda_L2*=0.42015.06 | 23.56 | |
| DADAData augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 24.17 | |
| Shake-Shake (26 2x96d)Backbone=Shake-Shake (26 2x96d), Augmentation=Cutout2018.05 | 24.2 | |
| Improved-GANData augmentation=false, Classifier architecture=13-layer CNN, Number of runs=32019.12 | 25.47 | |
| SWWAEdropout=false2015.11 | 27.83 | |
| SWWAEN=1000, Dropout=false, lambda_L2*=0.82015.06 | 27.83 | |
| Wide-ResNet-28-10Backbone=Wide-ResNet-28-10, Augmentation=Cutout2018.05 | 32.5 | |
| M1+M2Labeled samples=1000, Model type=Generative semi-supervised2014.06 | 36.02 | |
| M1+M22015.11 | 36.02 | |
| M1+M2N=10002015.06 | 36.02 | |
| M1+TSVMFeatures=M1 latent features, Kernel=RBF, Labeled samples=10002014.06 | 54.33 | |
| M1+TSVM2015.11 | 54.33 | |
| M1+TSVMN=10002015.06 | 54.33 | |
| M1+KNNFeatures=M1 latent features, Kernel=RBF, Labeled samples=10002014.06 | 65.63 | |
| M1+KNN2015.11 | 65.63 | |
| M1+KNNN=10002015.06 | 65.63 | |
| TSVMFeatures=Original features, Kernel=RBF, Labeled samples=10002014.06 | 66.55 | |
| TSVM2015.11 | 66.55 | |
| TSVMN=10002015.06 | 66.55 | |
| KNNFeatures=Original features, Kernel=RBF, Labeled samples=10002014.06 | 77.93 | |
| KNN2015.11 | 77.93 | |
| KNNN=10002015.06 | 77.93 |