Long-Tailed Image Classification on ImageNet-LT (test)
82Top-1 Acc (Overall)Ours
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| OursBackbone=ViT-B, Pre-training=DINOv2, Auxiliary data=true2024.10 | 82 | 84.7 | 81.5 | 76.2 | — | — | — | — | — | |
| Bal-CE†Backbone=ViT-B, Pre-training=DINOv2, Auxiliary data=true2024.10 | 80.5 | 83.8 | 79.8 | 73.4 | — | — | — | — | — | |
| Bal-CEBackbone=ViT-B, Pre-training=DINOv2, Auxiliary data=false2024.10 | 79.6 | 84.3 | 78.3 | 71.1 | — | — | — | — | — | |
| OursBackbone=ViT-B, Pre-training=CLIP, Auxiliary data=true2024.10 | 78.8 | 80.3 | 78.4 | 75.8 | — | — | — | — | — | |
| LIFT†Backbone=ViT-B, Pre-training=CLIP, Auxiliary data=true2024.10 | 77.8 | 80.2 | 77.2 | 73.1 | — | — | — | — | — | |
| LIFTBackbone=ViT-B, Pre-training=CLIP, Auxiliary data=false2024.10 | 77 | 80.2 | 76.1 | 71.5 | — | — | — | — | — | |
| BALLADBackbone=ResNet-50x16, #Epochs=50+102021.11 | 76.5 | 81.1 | 75.6 | 67 | — | — | — | — | — | |
| BALLADBackbone=ViT-B/16, #Epochs=50+102021.11 | 75.7 | 79.1 | 74.5 | 69.8 | — | — | — | — | — | |
| BALLADBackbone=ResNet-101, #Epochs=50+102021.11 | 70.5 | 74.7 | 69.1 | 63.3 | — | — | — | — | — | |
| OursBackbone=ViT-B, Pre-training=Scratch, Auxiliary data=true2024.10 | 68.2 | 74.5 | 66.2 | 57.4 | — | — | — | — | — | |
| BALLADBackbone=ResNet-50, #Epochs=50+102021.11 | 67.2 | 71 | 66.3 | 59.5 | — | — | — | — | — | |
| MDCSComponent=Ensemble, Metric Type=Diversity factor (σ) (%)2023.08 | 65.3 | 81.2 | 64.6 | 53.4 | — | — | — | — | — | |
| LiVT†Backbone=ViT-B, Pre-training=Scratch, Auxiliary data=true2024.10 | 65.2 | 73.7 | 62.8 | 49.8 | — | — | — | — | — | |
| MDCSComponent=Ensemble, Metric Type=Accuracy (%)2023.08 | 61.8 | 72.6 | 58.1 | 44.3 | — | — | — | — | — | |
| MDCSTraining Epochs=400, multi-experts=true, Backbone=ResNeXt-50, RandAug=true2023.08 | 61.8 | — | — | — | — | — | — | — | — | |
| SADEComponent=Ensemble, Metric Type=Diversity factor (σ) (%)2023.08 | 61.4 | 78.3 | 62.4 | 49.3 | — | — | — | — | — | |
| LiVTBackbone=ViT-B, Pre-training=Scratch, Auxiliary data=false2024.10 | 60.9 | 73.6 | 56.4 | 41 | — | — | — | — | — | |
| MDCSTraining Epochs=400, multi-experts=true, Backbone=ResNet-50, RandAug=true2023.08 | 60.7 | — | — | — | — | — | — | — | — | |
| NCLTraining Epochs=400, multi-experts=true, Backbone=ResNeXt-50, RandAug=true2023.08 | 60.5 | — | — | — | — | — | — | — | — | |
| RIDEComponent=Ensemble, Metric Type=Diversity factor (σ) (%)2023.08 | 60.2 | 76.6 | 62.9 | 51.8 | — | — | — | — | — | |
| MDCSTraining Epochs=180, multi-experts=true, Backbone=ResNeXt-50, RandAug=true2023.08 | 60.2 | — | — | — | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-101, #Epochs=4002021.11 | 60 | 68.2 | 58.7 | 41 | — | — | — | — | — | |
| NCLTraining Epochs=400, multi-experts=true, Backbone=ResNet-50, RandAug=true2023.08 | 59.5 | — | — | — | — | — | — | — | — | |
| MDCSTraining Epochs=180, multi-experts=true, Backbone=ResNet-50, RandAug=true2023.08 | 59.3 | — | — | — | — | — | — | — | — | |
| APA* + AGLUBackbone=SE-X502024.07 | 59.1 | 69.8 | 55.7 | 41.1 | — | — | — | — | — | |
| SADEComponent=Ensemble, Metric Type=Accuracy (%)2023.08 | 58.8 | 67 | 56.7 | 42.6 | — | — | — | — | — | |
| SADETraining Epochs=180, multi-experts=true, Backbone=ResNeXt-50, RandAug=false2023.08 | 58.8 | — | — | — | — | — | — | — | — | |
| APA*Backbone=SE-X502024.07 | 58.4 | 68.9 | 55.4 | 39.4 | — | — | — | — | — | |
| PaCoBackbone=ResNeXt-50, #Epochs=4002021.11 | 58.2 | 67.5 | 56.9 | 36.7 | — | — | — | — | — | |
| PaCoTraining Epochs=400, multi-experts=false, Backbone=ResNeXt-50, RandAug=true2023.08 | 58.2 | — | — | — | — | — | — | — | — | |
| Balanced SoftmaxBackbone=ResNeXt-101, #Epochs=4002021.11 | 58 | 69.2 | 55.8 | 36.3 | — | — | — | — | — | |
| APA* + AGLUBackbone=SE-R502024.07 | 57.9 | 68.3 | 54.8 | 39.4 | — | — | — | — | — | |
| GLMC+LossLoss reweighting strategy=Inverse-view reweighting strategy2026.05 | 57.6 | 67.2 | 55.6 | 38.8 | — | — | — | — | — | |
| GLMC+Loss+LRLoss reweighting strategy=Inverse-view reweighting strategy, Learning rate schedule=MiLeLR2026.05 | 57.6 | 67 | 55.9 | 39 | — | — | — | — | — | |
| PaCo + BatchFormerBackbone=ResNet-502022.03 | 57.4 | 62.7 | 56.7 | 42.1 | — | — | — | — | — | |
| APA*Backbone=SE-R502024.07 | 57.4 | 67.5 | 54.3 | 39.3 | — | — | — | — | — | |
| RIDE (3E)+CMO+CRBackbone=X502024.07 | 57.4 | 67.3 | 54.6 | 38.4 | — | — | — | — | — | |
| GLMC+SEL2026.05 | 57.2 | 68.7 | 54.4 | 38.3 | — | — | — | — | — | |
| MDCSComponent=Expert 2 (E2), Metric Type=Accuracy (%)2023.08 | 57.1 | 68.2 | 54.1 | 36.8 | — | — | — | — | — | |
| BCLTraining Epochs=180, multi-experts=false, Backbone=ResNeXt-50, RandAug=true2023.08 | 57.1 | — | — | — | — | — | — | — | — | |
| BCLBackbone=X502024.07 | 57.1 | 67.9 | 54.2 | 36.6 | — | — | — | — | — | |
| PaCoBackbone=ResNet-50, #Epochs=4002021.11 | 57 | 65 | 55.7 | 38.2 | — | — | — | — | — | |
| PaCoBackbone=ResNet-502022.03 | 57 | 64.8 | 55.9 | 39.1 | — | — | — | — | — | |
| PaCoTraining Epochs=400, multi-experts=false, Backbone=ResNet-50, RandAug=true2023.08 | 57 | — | — | — | — | — | — | — | — | |
| TSC+RIDEBackbone=ResNet-50, Number of experts=42021.11 | 56.9 | 69.2 | 52.4 | 37.9 | — | — | — | — | — | |
| RIDEBackbone=ResNeXt-50, #Epochs=1002021.11 | 56.8 | 68.2 | 53.8 | 36 | — | — | — | — | — | |
| RIDE (4E)Backbone=X502024.07 | 56.8 | 68.2 | 53.8 | 36 | — | — | — | — | — | |
| FeatRecon2026.05 | 56.8 | — | — | — | — | — | — | — | — | |
| Our BaselineBackbone=SE-X502024.07 | 56.7 | 67.9 | 53 | 37.7 | — | — | — | — | — | |
| ACE2022.03 | 56.6 | — | — | — | — | — | — | — | — | |
| ACETraining Epochs=400, multi-experts=true, Backbone=ResNeXt-50, RandAug=false2023.08 | 56.6 | — | — | — | — | — | — | — | — | |
| TSC+RIDEBackbone=ResNet-50, Number of experts=32021.11 | 56.3 | 69.1 | 51.7 | 36.7 | — | — | — | — | — | |
| RIDEComponent=Ensemble, Metric Type=Accuracy (%)2023.08 | 56.3 | 68 | 52.9 | 35.1 | — | — | — | — | — | |
| GLMC2026.05 | 56.3 | 70.1 | 52.4 | 30.4 | — | — | — | — | — | |
| Balanced SoftmaxBackbone=ResNeXt-50, #Epochs=4002021.11 | 56.2 | 67.7 | 53.8 | 34.2 | — | — | — | — | — | |
| BSCETraining Epochs=400, multi-experts=false, Backbone=ResNeXt-50, RandAug=true2023.08 | 56.2 | — | — | — | — | — | — | — | — | |
| RIDE (3E)+CMOBackbone=R502024.07 | 56.2 | 66.4 | 53.9 | 35.6 | — | — | — | — | — | |
| RIDE2022.03 | 56.1 | 67.9 | 52.3 | 36 | — | — | — | — | — | |
| ResLTBackbone=X502024.07 | 56.1 | 63.6 | 55.7 | 38.9 | — | — | — | — | — | |
| SSD2022.03 | 56 | 66.8 | 53.1 | 35.4 | — | — | — | — | — | |
| PaCoTraining Epochs=180, multi-experts=false, Backbone=ResNeXt-50, RandAug=true2023.08 | 56 | — | — | — | — | — | — | — | — | |
| τ-normTraining Epochs=400, multi-experts=false, Backbone=ResNeXt-50, RandAug=false2023.08 | 56 | — | — | — | — | — | — | — | — | |
| Our BaselineBackbone=SE-R502024.07 | 56 | 66.2 | 53.1 | 37.1 | — | — | — | — | — | |
| SSDBackbone=X502024.07 | 56 | 66.8 | 53.1 | 35.4 | — | — | — | — | — | |
| TSC+RIDEBackbone=ResNet-50, Number of experts=22021.11 | 55.9 | 68.4 | 51.3 | 36.4 | — | — | — | — | — | |
| RIDE-3e + BatchFormerBackbone=ResNet-50, Training Stage=two-stage2022.03 | 55.7 | 64.6 | 53.4 | 39 | — | — | — | — | — | |
| MDCSComponent=Expert 3 (E3), Metric Type=Accuracy (%)2023.08 | 55.7 | 51.8 | 56.5 | 58.8 | — | — | — | — | — | |
| RIDEBackbone=ResNet-50, #Epochs=1002021.11 | 55.4 | 66.2 | 52.3 | 36.5 | — | — | — | — | — | |
| RIDEBackbone=ResNet-50, Number of experts=42021.11 | 55.4 | 66.2 | 52.3 | 36.5 | — | — | — | — | — | |
| LCRegBackbone=ResNet-502022.06 | 55.3 | — | — | — | — | — | — | — | — | |
| LWS+ImbSAMBackbone=X502024.07 | 55.3 | 63.2 | 53.7 | 38.3 | — | — | — | — | — | |
| ResLTBackbone=ResNeXt-101, #Epochs=1802021.11 | 55.1 | 63.3 | 53.3 | 40.3 | — | — | — | — | — | |
| TLCexperts=42021.11 | 55.1 | 68.9 | 55.7 | 40.8 | — | — | — | — | — | |
| OPeNBackbone=ResNeXt-502021.12 | 55.1 | — | — | — | — | — | — | — | — | |
| Balanced SoftmaxBackbone=ResNet-50, #Epochs=4002021.11 | 55 | 66.7 | 52.9 | 33 | — | — | — | — | — | |
| BSCETraining Epochs=400, multi-experts=false, Backbone=ResNet-50, RandAug=true2023.08 | 55 | — | — | — | — | — | — | — | — | |
| DOCBackbone=R502024.07 | 55 | 65.1 | 52.8 | 34.2 | — | — | — | — | — | |
| RIDE-3eBackbone=ResNet-50, Training Stage=two-stage2022.03 | 54.9 | 66.2 | 51.7 | 34.9 | — | — | — | — | — | |
| RIDEBackbone=ResNet-50, Number of experts=32021.11 | 54.9 | 66.2 | 51.7 | 34.9 | — | — | — | — | — | |
| RIDE-3 experts2026.05 | 54.9 | 66.2 | 51.7 | 34.9 | — | — | — | — | — | |
| ACETraining Epochs=400, multi-experts=true, Backbone=ResNet-50, RandAug=false2023.08 | 54.7 | — | — | — | — | — | — | — | — | |
| RIDE2021.11 | 54.6 | 70.6 | 54.8 | 38.3 | — | — | — | — | — | |
| TLCexperts=32021.11 | 54.6 | 69.3 | 56.7 | 37.9 | — | — | — | — | — | |
| τ-normTraining Epochs=400, multi-experts=false, Backbone=ResNet-50, RandAug=false2023.08 | 54.5 | — | — | — | — | — | — | — | — | |
| PaCo2022.03 | 54.4 | 63.2 | 51.6 | 39.2 | — | — | — | — | — | |
| RIDEBackbone=ResNet-50, Number of experts=22021.11 | 54.4 | 65.8 | 51 | 34.6 | — | — | — | — | — | |
| Focal-SAM2026.05 | 54.3 | 63.9 | 52.2 | 34.4 | — | — | — | — | — | |
| CNTBackbone=X502024.07 | 54.2 | 63.2 | 52.1 | 36.9 | — | — | — | — | — | |
| TLCexperts=22021.11 | 54.1 | 68.7 | 55.4 | 38.3 | — | — | — | — | — | |
| RIDE-3e + BatchFormerBackbone=ResNet-50, Training Stage=one-stage2022.03 | 54.1 | 64.3 | 51.4 | 35.1 | — | — | — | — | — | |
| Weight Balancingvariant=WD + WD & Max2022.03 | 53.9 | 62.5 | 50.4 | 41.5 | — | — | — | — | — | |
| SADEComponent=Expert 2 (E2), Metric Type=Accuracy (%)2023.08 | 53.9 | 65.5 | 50.5 | 33.3 | — | — | — | — | — | |
| INC-DRW2026.05 | 53.9 | 67.1 | 49.7 | 29 | — | — | — | — | — | |
| MisLASBackbone=ResNeXt-50, Source=Reproduced by authors2021.12 | 53.7 | — | — | — | — | — | — | — | — | |
| RIDE-3eBackbone=ResNet-50, Training Stage=one-stage2022.03 | 53.6 | 64.9 | 50.4 | 33.2 | — | — | — | — | — | |
| DRO-LT2022.03 | 53.5 | 64 | 49.8 | 33.1 | — | — | — | — | — | |
| LWSBackbone=ResNeXt-152, #Epochs=902021.11 | 53.3 | 63.5 | 50.4 | 34.2 | — | — | — | — | — | |
| Weight Balancingvariant=WD + WD2022.03 | 53.3 | 62 | 49.7 | 41 | — | — | — | — | — | |
| ALABackbone=X502024.07 | 53.3 | 64.1 | 49.9 | 34.7 | — | — | — | — | — | |
| RBL2026.05 | 53.3 | 64.8 | 49.6 | 34.2 | — | — | — | — | — |