Image Classification on CIFAR-100-LT (Imbalance Ratio 50)
91.2Top-1 AccuracyGNM-PT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GNM-PTBackbone=ViT-B/16, Optimization Strategy (SAM)=false2024.10 | 91.2 | — | |
| LPTBackbone=ViT-B/16, Optimization Strategy (SAM)=false2024.10 | 90 | — | |
| VPTBackbone=ViT-B/16, Optimization Strategy (SAM)=false2024.10 | 84.8 | — | |
| SHIKEBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 59.8 | — | |
| ImbSAMBackbone=ResNet32, Optimization Strategy (SAM)=true2024.10 | 59.3 | — | |
| NCLBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 58.2 | — | |
| DRO-LT (copied)Training setting=decoupled learning, Backbone=ResNet-32, Source=Copied from original paper2022.09 | 57.57 | — | |
| Difficulty-Net + LASTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 56.9 | — | |
| Difficulty-Net + Balanced SoftmaxTraining setting=e2e training, Backbone=ResNet-322022.09 | 56.86 | — | |
| Difficulty-Net + LWSTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 56.61 | — | |
| BCLBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 56.6 | — | |
| GPaCotrained in same setting=true2022.09 | 56.4 | — | |
| GPaCoBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 56.4 | — | |
| Difficulty-Net + cRTTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 56.34 | — | |
| PaCo + Balanced Softmax (copied)Training setting=e2e training, Backbone=ResNet-32, Source=Copied from original paper2022.09 | 56 | — | |
| PaCotrained in same setting=true2022.09 | 56 | — | |
| PaCo + Balanced SoftmaxTraining setting=e2e training, Backbone=ResNet-322022.09 | 55.88 | — | |
| RIDEBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 55 | — | |
| BALMSTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 54.42 | — | |
| Difficulty-NetTraining setting=e2e training, Backbone=ResNet-322022.09 | 54.27 | — | |
| Balanced Softmaxtrained in same setting=true2022.09 | 54.2 | — | |
| MWN + LASTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.95 | — | |
| CCSAMBackbone=ResNet32, Optimization Strategy (SAM)=true2024.10 | 53.9 | — | |
| LASTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.85 | — | |
| MWN + LWSTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.77 | — | |
| MWN + cRTTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.62 | — | |
| GCLBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 53.6 | — | |
| LWSTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.44 | — | |
| cRTTraining setting=decoupled learning, Backbone=ResNet-322022.09 | 53.32 | — | |
| RIDE (3 experts) + CMOBackbone=ResNet-32, CMO=true, Number of experts=32021.12 | 53 | — | |
| MISLASBackbone=ResNet-32, CMO=false2021.12 | 52.3 | — | |
| MISLAS2022.09 | 52.3 | — | |
| MisLASBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 52.3 | — | |
| PaCoTraining setting=e2e training, Backbone=ResNet-322022.09 | 52.14 | — | |
| LDAM-DRW + CMOBackbone=ResNet-32, CMO=true2021.12 | 51.7 | — | |
| RIDE (3 experts)Backbone=ResNet-32, CMO=false, Number of experts=32021.12 | 51.4 | — | |
| BS + CMOBackbone=ResNet-32, CMO=true2021.12 | 51.4 | — | |
| EQLTraining setting=e2e training, Backbone=ResNet-322022.09 | 51.34 | — | |
| CE-DRW + CMOBackbone=ResNet-32, CMO=true2021.12 | 50.9 | — | |
| LADEBackbone=ResNet-32, CMO=false2021.12 | 50.5 | — | |
| Causal NormBackbone=ResNet-32, CMO=false2021.12 | 50.3 | — | |
| Causal Norm2022.09 | 50.3 | — | |
| ResLT2022.09 | 50 | — | |
| Balanced Softmax (BS)Backbone=ResNet-32, CMO=false2021.12 | 49.9 | — | |
| CB-DATraining setting=e2e training, Backbone=ResNet-322022.09 | 49.8 | — | |
| MWNTraining setting=e2e training, Backbone=ResNet-322022.09 | 49.68 | — | |
| RemixBackbone=ResNet-32, CMO=false2021.12 | 49.5 | — | |
| CDB-CETraining setting=e2e training, Backbone=ResNet-322022.09 | 49.45 | — | |
| LDAMTraining setting=e2e training, Backbone=ResNet-322022.09 | 49.19 | — | |
| Focal LossTraining setting=e2e training, Backbone=ResNet-322022.09 | 48.91 | — | |
| IB LossBackbone=ResNet-32, CMO=false2021.12 | 48.9 | — | |
| CE + CMOBackbone=ResNet-32, CMO=true2021.12 | 48.3 | — | |
| LDAM-DRWBackbone=ResNet-32, CMO=false2021.12 | 47.9 | — | |
| Class-BalancedTraining setting=e2e training, Backbone=ResNet-322022.09 | 47.67 | — | |
| BBNBackbone=ResNet-32, CMO=false2021.12 | 47.1 | — | |
| BBN2022.09 | 47 | — | |
| BBNBackbone=ResNet32, Optimization Strategy (SAM)=false2024.10 | 47 | — | |
| LDAM+DRW2022.09 | 46.6 | — | |
| CE-DRWBackbone=ResNet-32, CMO=false2021.12 | 45.6 | — | |
| Focal Loss2022.09 | 44.3 | — | |
| Cross Entropy (CE)Backbone=ResNet-32, CMO=false2021.12 | 44 | — | |
| Balanced Softmax2020.12 | — | 49.9 | |
| BBN2020.12 | — | 47 | |
| Causal Norm2020.12 | — | 50.3 | |
| Focal Loss2020.12 | — | 44.3 | |
| LADE2020.12 | — | 50.5 | |
| LDAM2020.12 | — | 46.6 | |
| PC Softmax2020.12 | — | 49.5 | |
| Softmax2020.12 | — | 45.5 |