Image Classification on CIFAR-10-LT (test)
0.0824Top-1 ErrorLDAM-DRW + Du @5x
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.0824 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.0868 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.0872 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=2p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.0897 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1022 | — | |
| MetaSAug with LDAM lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1032 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1048 | — | |
| MetaSAug with cross-entropy lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1056 | — | |
| MetaSAug with focal lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1074 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1086 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=10, Unlabeled Imbalance Ratio (pu)=2p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1104 | — | |
| Meta-class-weight with cross-entropy lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1115 | — | |
| Meta-class-weight with focal lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1163 | — | |
| LDAM-DRWImbalance factor=10, Backbone=ResNet-322021.03 | 0.1163 | — | |
| BBNImbalance factor=10, Backbone=ResNet-322021.03 | 0.1168 | — | |
| LDAM-DRWLabeled Imbalance Ratio (p)=10, Backbone=ResNet-32, Amount of Unlabeled Data=None2020.06 | 0.1184 | — | |
| MetaSAug with LDAM lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.119 | — | |
| MetaSAug with LDAM lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.119 | — | |
| MetaSAug with cross-entropy lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1236 | — | |
| MetaSAug with CE lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1236 | — | |
| Meta-weight netImbalance factor=10, Backbone=ResNet-322021.03 | 0.1245 | — | |
| Optimal Transport Re-weighting (Weight Vector)Backbone=ResNet-32, Imbalance Factor=202022.08 | 0.125 | — | |
| Class-balanced focal lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1252 | — | |
| Meta-class-weight with LDAM lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.126 | — | |
| LDAM lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1268 | — | |
| MetaSAug with focal lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1284 | — | |
| MetaSAug with focal lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1284 | — | |
| MixupImbalance factor=10, Backbone=ResNet-322021.03 | 0.129 | — | |
| Class-balanced cross-entropy lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.131 | — | |
| Focal lossImbalance factor=10, Backbone=ResNet-322021.03 | 0.1334 | — | |
| Meta-class-weight with cross-entropy lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1354 | — | |
| Meta-class-weight with CE lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1354 | — | |
| CELabeled Imbalance Ratio (p)=10, Backbone=ResNet-32, Amount of Unlabeled Data=None2020.06 | 0.1361 | — | |
| Cross-entropy trainingImbalance factor=10, Backbone=ResNet-322021.03 | 0.1382 | — | |
| Meta-class-weight with focal lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.139 | — | |
| Meta-class-weight with focal lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.139 | — | |
| IB + Focal lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1432 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1433 | — | |
| IBBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1459 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.147 | — | |
| IB + CBBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1473 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1493 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1496 | — | |
| LDAM-DRWImbalance factor=20, Backbone=ResNet-322021.03 | 0.151 | — | |
| LDAM-DRWBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.151 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1518 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=2p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1524 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1533 | — | |
| Class-balanced cross-entropy lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1544 | — | |
| CB, CE lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1544 | — | |
| Optimal Transport Re-weighting (Weight Vector)Backbone=ResNet-32, Imbalance Factor=502022.08 | 0.1554 | — | |
| LDAM-DRW + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=2p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1555 | — | |
| Meta-weight netImbalance factor=20, Backbone=ResNet-322021.03 | 0.1555 | — | |
| Meta-weight netBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1555 | — | |
| Meta-class-weight with LDAM lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1563 | — | |
| Meta-class-weight with LDAM lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1563 | — | |
| MetaSAug with LDAM lossImbalance factor=50, Backbone=ResNet-322021.03 | 0.1566 | — | |
| MetaSAug with LDAM lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1566 | — | |
| MetaSAug with focal lossImbalance factor=50, Backbone=ResNet-322021.03 | 0.1596 | — | |
| MetaSAug with focal lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1596 | — | |
| MetaSAug with cross-entropy lossImbalance factor=50, Backbone=ResNet-322021.03 | 0.1597 | — | |
| MetaSAug with CE lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1597 | — | |
| LDAM lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1611 | — | |
| LDAM lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1611 | — | |
| Class-balanced focal lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1622 | — | |
| CB, Focal lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1622 | — | |
| Class-balanced fine-tuningImbalance factor=20, Backbone=ResNet-322021.03 | 0.1678 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1679 | — | |
| Class-balanced fine-tuningImbalance factor=10, Backbone=ResNet-322021.03 | 0.1683 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1688 | — | |
| L2RWBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.169 | — | |
| Meta-class-weight with focal lossImbalance factor=50, Backbone=ResNet-322021.03 | 0.1712 | — | |
| Meta-class-weight with focal lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1712 | — | |
| Focal lossImbalance factor=20, Backbone=ResNet-322021.03 | 0.1724 | — | |
| Focal lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1724 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=1, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1748 | — | |
| Cross-entropy trainingImbalance factor=20, Backbone=ResNet-322021.03 | 0.1756 | — | |
| CE lossBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1756 | — | |
| IB + Focal lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1762 | — | |
| Meta-class-weight with LDAM lossImbalance factor=50, Backbone=ResNet-322021.03 | 0.1777 | — | |
| Meta-class-weight with LDAM lossBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1777 | — | |
| BBNImbalance factor=50, Backbone=ResNet-322021.03 | 0.1782 | — | |
| BBNBackbone=ResNet-32, Imbalance Factor=202022.08 | 0.1782 | — | |
| L2RWImbalance factor=10, Backbone=ResNet-322021.03 | 0.1788 | — | |
| IB + CBBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1791 | — | |
| Optimal Transport Re-weighting (Weight Vector)Backbone=ResNet-32, Imbalance Factor=1002022.08 | 0.1813 | — | |
| IBBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1834 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=50, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1836 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=p/2, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1842 | — | |
| L2RWImbalance factor=20, Backbone=ResNet-322021.03 | 0.1865 | — | |
| LDAM-DRWImbalance factor=50, Backbone=ResNet-322021.03 | 0.1873 | — | |
| LDAM-DRWBackbone=ResNet-32, Imbalance Factor=502022.08 | 0.1873 | — | |
| CE + Du @5xLabeled Imbalance Ratio (p)=100, Unlabeled Imbalance Ratio (pu)=p, Backbone=ResNet-32, Amount of Unlabeled Data=5x2020.06 | 0.1874 | — | |
| LDAM-DRWLabeled Imbalance Ratio (p)=50, Backbone=ResNet-32, Amount of Unlabeled Data=None2020.06 | 0.1906 | — | |
| MetaSAug with LDAM lossImbalance factor=100, Backbone=ResNet-322021.03 | 0.1934 | — | |
| MetaSAug with LDAM lossBackbone=ResNet-32, Imbalance Factor=1002022.08 | 0.1934 | — | |
| MetaSAug with focal lossImbalance factor=100, Backbone=ResNet-322021.03 | 0.1936 | — | |
| MetaSAug with focal lossBackbone=ResNet-32, Imbalance Factor=1002022.08 | 0.1936 | — | |
| MetaSAug with cross-entropy lossImbalance factor=100, Backbone=ResNet-322021.03 | 0.1946 | — | |
| MetaSAug with CE lossBackbone=ResNet-32, Imbalance Factor=1002022.08 | 0.1946 | — |