Image Classification on ImageNet-LT 1.0 (test)
58.5Top-1 AccuracyBCL + ConCutMix
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BCL + ConCutMixBackbone=ResNet-50, Training Epochs=1002024.07 | 58.5 | — | — | — | — | |
| BCL + CMOBackbone=ResNet-50, Training Epochs=1002024.07 | 57.3 | — | — | — | — | |
| BCL + CutMixBackbone=ResNet-50, Training Epochs=1002024.07 | 57.1 | — | — | — | — | |
| RIDE (4 experts)Backbone=ResNeXt-50, GFlops=5.19 (1.2x)2020.10 | 56.8 | — | — | — | — | |
| SMCBackbone=ResNet-50, Training Epochs=1002024.07 | 56.6 | — | — | — | — | |
| RIDE (3 experts)Backbone=ResNeXt-50, GFlops=4.69 (1.1x)2020.10 | 56.4 | — | — | — | — | |
| BCL + MixupBackbone=ResNet-50, Training Epochs=1002024.07 | 56.3 | — | — | — | — | |
| BCLBackbone=ResNet-50, Training Epochs=1002024.07 | 55.9 | — | — | — | — | |
| RIDE (2 experts)Backbone=ResNeXt-50, GFlops=3.92 (0.9x)2020.10 | 55.9 | — | — | — | — | |
| RIDE (4 experts)Backbone=ResNet-50, GFlops=5.15 (1.3x)2020.10 | 55.4 | — | — | — | — | |
| RIDE (3 experts)Backbone=ResNet-50, GFlops=4.36 (1.1x)2020.10 | 54.9 | — | — | — | — | |
| RIDE (2 experts)Backbone=ResNet-50, Training Epochs=1002024.07 | 54.4 | — | — | — | — | |
| RIDE (2 experts)Backbone=ResNet-50, GFlops=3.71 (0.9x)2020.10 | 54.4 | — | — | — | — | |
| DisalignBackbone=ResNet-50, Training Epochs=1002024.07 | 52.9 | — | — | — | — | |
| KCLBackbone=ResNet-50, Training Epochs=1002024.07 | 52.4 | — | — | — | — | |
| TSCBackbone=ResNet-50, Training Epochs=1002024.07 | 52.4 | — | — | — | — | |
| SSPBackbone=ResNet-50, Training Epochs=1002024.07 | 51.3 | — | — | — | — | |
| FBLBackbone Net=ResNet-502023.05 | 50.7 | — | — | — | — | |
| LABackbone Net=ResNet-502023.05 | 50.44 | — | — | — | — | |
| LWSBackbone=ResNet-50, Training Epochs=1002024.07 | 49.9 | — | — | — | — | |
| LWSBackbone=ResNeXt-50, GFlops=4.26 (1.0x)2020.10 | 49.9 | — | — | — | — | |
| cRTBackbone=ResNet-50, Training Epochs=1002024.07 | 49.6 | — | — | — | — | |
| cRTBackbone=ResNeXt-50, GFlops=4.26 (1.0x)2020.10 | 49.6 | — | — | — | — | |
| tau-normBackbone=ResNeXt-50, GFlops=4.26 (1.0x)2020.10 | 49.4 | — | — | — | — | |
| LDAM-DRWBackbone Net=ResNet-502023.05 | 48.8 | — | — | — | — | |
| UniMixBackbone=ResNet-50, Training Epochs=1002024.07 | 48.4 | — | — | — | — | |
| DecouplingBackbone Net=ResNet-502023.05 | 47.7 | — | — | — | — | |
| LWSBackbone=ResNet-50, GFlops=4.11 (1.0x)2020.10 | 47.7 | — | — | — | — | |
| cRTBackbone=ResNet-50, GFlops=4.11 (1.0x)2020.10 | 47.3 | — | — | — | — | |
| NCMBackbone=ResNeXt-50, GFlops=4.26 (1.0x)2020.10 | 47.3 | — | — | — | — | |
| τ-normBackbone=ResNet-50, Training Epochs=1002024.07 | 46.7 | — | — | — | — | |
| tau-normBackbone=ResNet-50, GFlops=4.11 (1.0x)2020.10 | 46.7 | — | — | — | — | |
| OLTRBackbone=ResNeXt-502020.10 | 46.3 | — | — | — | — | |
| CE lossBackbone Net=ResNet-502023.05 | 44.51 | — | — | — | — | |
| Cross Entropy (CE)Backbone=ResNeXt-50, GFlops=4.26 (1.0x)2020.10 | 44.4 | — | — | — | — | |
| NCMBackbone=ResNet-50, GFlops=4.11 (1.0x)2020.10 | 44.3 | — | — | — | — | |
| Cross Entropy (CE)Backbone=ResNet-50, GFlops=4.11 (1.0x)2020.10 | 41.6 | — | — | — | — | |
| AutoBalanceBackbone=ResNet-50, Design Space=τ, Loss Function=LA loss2022.01 | — | — | — | — | 52.39 | |
| AutoBalanceBackbone=ResNet-50, Design Space=l2022.01 | — | — | — | — | 52.6 | |
| AutoBalanceBackbone=ResNet-50, Design Space=Δ2022.01 | — | — | — | — | 53.02 | |
| AutoBalanceBackbone=ResNet-50, Design Space=Δ & l2022.01 | — | — | — | — | 51.74 | |
| AutoBalanceBackbone=ResNet-50, Design Space=Δ & l, Initialization=LA loss (τ=1)2022.01 | — | — | — | — | 50.91 | |
| CDT lossBackbone=ResNet-502022.01 | — | — | — | — | 53.47 | |
| Class-balanced Cross-Entropy (CB CE)Backbone=ResNet-322020.03 | — | 73.41 | 59.22 | 50.49 | — | |
| Cross-EntropyBackbone=ResNet-502022.01 | — | — | — | — | 55.47 | |
| Cross-Entropy (CE)Backbone=ResNet-322020.03 | — | 74.74 | 61.35 | 52.12 | — | |
| LA lossBackbone=ResNet-50, Temperature (tau)=12022.01 | — | — | — | — | 52.46 | |
| LDAM lossBackbone=ResNet-502022.01 | — | — | — | — | 54.21 | |
| Meta-learning for long-tailed recognition (Ours, CE)Backbone=ResNet-322020.03 | — | 70.1 | 53.29 | 45.18 | — |