Domain Adaptation on MNIST to MNIST-M (test)
99.4Acc (Target)SBADA-GAN
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| SBADA-GANClassifier training space=Weighted combination2017.05 | 99.4 | — | — | — | — | — | |
| SBADA-GANClassifier training space=Target domain (Ct)2017.05 | 99.1 | — | — | — | — | — | |
| SBADA-GANClassifier training space=Source domain (Cs)2017.05 | 98.4 | — | — | — | — | — | |
| PixelDA2017.05 | 98.2 | — | — | — | — | — | |
| Target Only2017.05 | 96.4 | — | — | — | — | — | |
| EmaQW/A bit=4/4, Backbone=VGG-52026.06 | 96.1 | — | — | — | — | — | |
| EmaQW/A bit=3/3, Backbone=VGG-52026.06 | 95.8 | — | — | — | — | — | |
| EmaQW/A bit=2/2, Backbone=VGG-52026.06 | 95.5 | — | — | — | — | — | |
| EmaQW/A bit=32/32, Backbone=VGG-52026.06 | 95.3 | — | — | — | — | — | |
| Asymmetric Tri-trainingWeight constraint=lambda = 02017.02 | 94.2 | — | — | — | — | — | |
| ATT2017.05 | 94.2 | — | — | — | — | — | |
| Asymmetric Tri-training2017.02 | 94 | — | — | — | — | — | |
| APoTW/A bit=4/4, Backbone=VGG-52026.06 | 91.6 | — | — | — | — | — | |
| DANNW/A bit=32/32, Backbone=VGG-52026.06 | 91.3 | — | — | — | — | — | |
| DoReFaW/A bit=4/4, Backbone=VGG-52026.06 | 90.6 | — | — | — | — | — | |
| DA_assParameter tuning=Fixed2017.05 | 89.5 | — | — | — | — | — | |
| DA_assParameter tuning=Tuned2017.05 | 89.5 | — | — | — | — | — | |
| DoReFaW/A bit=3/3, Backbone=VGG-52026.06 | 88.5 | — | — | — | — | — | |
| ZAQW/A bit=4/4, Backbone=VGG-52026.06 | 88.3 | — | — | — | — | — | |
| Choi et al.W/A bit=4/4, Backbone=VGG-52026.06 | 87.4 | — | — | — | — | — | |
| KNN-Ad2017.02 | 86.7 | — | — | — | — | — | |
| TRUDA2017.05 | 86.7 | — | — | — | — | — | |
| ZeroQW/A bit=4/4, Backbone=VGG-52026.06 | 86.6 | — | — | — | — | — | |
| Asymmetric Tri-trainingBatch Normalization=false2017.02 | 85.3 | — | — | — | — | — | |
| APoTW/A bit=3/3, Backbone=VGG-52026.06 | 85.2 | — | — | — | — | — | |
| DoReFaW/A bit=2/2, Backbone=VGG-52026.06 | 83.5 | — | — | — | — | — | |
| DSN2017.02 | 83.2 | — | — | — | — | — | |
| DSN2017.05 | 83.2 | — | — | — | — | — | |
| LLSQW/A bit=4/4, Backbone=VGG-52026.06 | 81.8 | — | — | — | — | — | |
| DANN2017.02 | 81.5 | — | — | — | — | — | |
| LLSQW/A bit=3/3, Backbone=VGG-52026.06 | 80.9 | — | — | — | — | — | |
| Choi et al.W/A bit=3/3, Backbone=VGG-52026.06 | 77.5 | — | — | — | — | — | |
| DANN2017.05 | 77.4 | — | — | — | — | — | |
| MMD2017.02 | 76.9 | — | — | — | — | — | |
| MMD2017.05 | 76.9 | — | — | — | — | — | |
| ZeroQW/A bit=3/3, Backbone=VGG-52026.06 | 76.9 | — | — | — | — | — | |
| ZAQW/A bit=3/3, Backbone=VGG-52026.06 | 66.8 | — | — | — | — | — | |
| Source Only2017.05 | 63.6 | — | — | — | — | — | |
| CORAL2017.05 | 63.1 | — | — | — | — | — | |
| CoGAN2017.05 | 62 | — | — | — | — | — | |
| Source Only w/o BNBatch Normalization=false2017.02 | 59.1 | — | — | — | — | — | |
| Source onlyW/A bit=32/32, Backbone=VGG-52026.06 | 58.8 | — | — | — | — | — | |
| Source Only with BNBatch Normalization=true2017.02 | 57.1 | — | — | — | — | — | |
| LLSQW/A bit=2/2, Backbone=VGG-52026.06 | 57.1 | — | — | — | — | — | |
| LSQW/A bit=4/4, Backbone=VGG-52026.06 | 55.5 | — | — | — | — | — | |
| LSQW/A bit=3/3, Backbone=VGG-52026.06 | 54.6 | — | — | — | — | — | |
| LSQW/A bit=2/2, Backbone=VGG-52026.06 | 52.7 | — | — | — | — | — | |
| Associative Domain AdaptationHyper-parameter optimization=Individual2017.08 | — | — | — | — | 10.53 | 85.94 | |
| Baseline (N/A)backbone=ResNet [68], batch size=128, samples in max step (N')=50002023.02 | — | — | — | 0 | — | — | |
| CORAL2017.08 | — | — | — | — | 42.3 | — | |
| DAMMDDomain adaptation setting=DAMMD (L_assoc replaced with MMD)2017.08 | — | — | — | — | 22.9 | — | |
| DANN2017.08 | — | — | — | — | 23.33 | 55.87 | |
| DSN w/ DANN2017.08 | — | — | — | — | 16.8 | 63.18 | |
| DSN w/ MMD2017.08 | — | — | — | — | 19.5 | 56.77 | |
| MMD2017.08 | — | — | — | — | 23.1 | — | |
| Neural TCbackbone=ResNet [68], batch size=128, samples in max step (N')=50002023.02 | — | — | — | 3.282 | — | — | |
| SA2017.08 | — | — | — | — | 43.1 | — | |
| Slicebackbone=ResNet [68], batch size=128, samples in max step (N')=50002023.02 | — | — | — | 2.578 | — | — | |
| Source only2017.08 | — | — | — | — | 35.96 | — | |
| Target only2017.08 | — | — | — | — | 6.37 | — | |
| Transf. Repr.2017.08 | — | — | — | — | 13.3 | — | |
| vCLUBbackbone=ResNet [68], batch size=128, samples in max step (N')=50002023.02 | — | — | — | 3.123 | — | — |