Zero-Shot Domain Generalization on DomainNet unseen domains
68.27Clipart AccuracyBrAD
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BrADLabel Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 68.27 | 26.6 | 31.08 | 34.03 | 48.17 | 38.74 | 38.48 | |
| BrADLabel Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 64.01 | 25.02 | 29.32 | 29.64 | 44.09 | 35.37 | 34.95 | |
| BrADLabel Fraction=10%, Evaluation Protocol=kNN, Backbone=ResNet182021.12 | 60.78 | 19.76 | 26.06 | 31.56 | 41.38 | 34.77 | 37.43 | |
| BrADLabel Fraction=5%, Evaluation Protocol=kNN, Backbone=ResNet182021.12 | 55.75 | 18.15 | 24.29 | 26.93 | 37.54 | 31.12 | 33.33 | |
| BrADLabel Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 47.26 | 16.89 | 20.03 | 23.74 | 31.67 | 25.85 | 25.08 | |
| BrADLabel Fraction=1%, Evaluation Protocol=kNN, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 40.65 | 14 | 16.8 | 21.28 | 25.72 | 22.35 | 22.29 | |
| DIULLabel Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 39.32 | 19.09 | 21.09 | 10.5 | 28.49 | 23.31 | 30.51 | |
| SimCLR V2Label Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 37.11 | 19.87 | 24.01 | 12.33 | 31.58 | 24.28 | 30.17 | |
| DIULLabel Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 35.15 | 20.88 | 25.9 | 15.69 | 30.77 | 26.09 | 33.29 | |
| SimCLR V2Label Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 34.03 | 17.17 | 21.35 | 10.88 | 27.46 | 20.89 | 24.34 | |
| MoCo V2Label Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 32.46 | 18.54 | 25.35 | 8.05 | 23.71 | 21.87 | 29.91 | |
| AdCoLabel Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 32.25 | 17.96 | 23.35 | 11.56 | 27.57 | 22.79 | 29.98 | |
| AdCoLabel Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 30.77 | 18.65 | 19.97 | 7.75 | 24.19 | 19.42 | 24.31 | |
| MoCo V2Label Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 28.13 | 13.79 | 20.8 | 9.67 | 21.44 | 18.99 | 24.91 | |
| CuMixBackbone=ResNet-502020.07 | 27.6 | 17.8 | 25.5 | 9.9 | 22.6 | 20.7 | — | |
| mixup+SPNetBackbone=ResNet-502020.07 | 27.2 | 16.9 | 24.7 | 8.5 | 21.3 | 19.7 | — | |
| Epi-FCR+SPNetBackbone=ResNet-502020.07 | 26.4 | 16.7 | 24.6 | 9.2 | 23.2 | 20 | — | |
| SPNetBackbone=ResNet-502020.07 | 26 | 16.9 | 23.8 | 8.2 | 21.8 | 19.4 | — | |
| SimCLR V2Label Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 23.51 | 15.42 | 20.25 | 5.29 | 18.85 | 15.46 | 17.84 | |
| MoCo V2Label Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 18.85 | 10.57 | 11.38 | 6.32 | 15.28 | 12.12 | 14.97 | |
| DIULLabel Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 18.53 | 10.62 | 14.45 | 12.65 | 21.3 | 16.56 | 21.68 | |
| AdCoLabel Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 16.16 | 12.26 | 11.13 | 5.65 | 17.19 | 12.47 | 16.53 | |
| ERMLabel Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 15.1 | 9.39 | 9.9 | 7.11 | 5.12 | 8.94 | 9.19 | |
| BYOLLabel Fraction=10%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 14.55 | 8.71 | 9.5 | 5.95 | 4.45 | 8.69 | 10.38 | |
| ERMLabel Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 10.21 | 7.08 | 7.45 | 5.34 | 5 | 6.5 | 6.08 | |
| BYOLLabel Fraction=5%, Evaluation Protocol=Linear classifier, Backbone=ResNet182021.12 | 9.6 | 5.09 | 9.78 | 6.02 | 3.97 | 7.83 | 10.73 | |
| ERMLabel Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training=Randomly initialized2021.12 | 6.54 | 2.96 | 6.68 | 5 | 7.25 | 5.88 | 6.97 | |
| BYOLLabel Fraction=1%, Evaluation Protocol=Linear classifier, Backbone=ResNet18, Pre-training Epochs=10002021.12 | 6.21 | 3.48 | 5 | 4.27 | 4.42 | 5.61 | 8.47 |