Long-tailed Image Classification on CIFAR-100-LT λ=100 (test)
89.1Accuracy (top-1)LPT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LPTtraining_protocol=Vision-only Pretrained2022.10 | 89.1 | — | |
| VPTtraining_protocol=Vision-only Pretrained2022.10 | 81 | — | |
| BALLAD (Ma et al., 2021)training_protocol=VL Pretrained with Extra Data2022.10 | 77.8 | — | |
| PaCotraining_protocol=Training from Scratch2022.10 | 52 | — | |
| Zhu et al. (2022)training_protocol=Training from Scratch2022.10 | 51.9 | — | |
| Li et al. (2022)training_protocol=Training from Scratch2022.10 | 48.7 | — | |
| ATL_all2022.12 | 47.11 | — | |
| ATL_ALAS2022.12 | 46.54 | — | |
| MiSLASTraining Stage=Two-stage2022.12 | 46.05 | — | |
| L2ATraining Stage=Two-stage2022.12 | 45.21 | — | |
| TDE2022.12 | 44.1 | — | |
| ABS2022.12 | 43.88 | — | |
| TSC2022.12 | 43.8 | — | |
| RCBM-CE2022.12 | 43.3 | — | |
| BBN2022.12 | 42.5 | — | |
| ATL_AEM2022.12 | 40.77 | — | |
| L2RW2022.12 | 40.2 | — | |
| L2ATraining Stage=One-stage2022.12 | 39.83 | — | |
| CB Loss2022.12 | 39.6 | — | |
| MiSLASTraining Stage=One-stage2022.12 | 38.87 | — | |
| Focal Loss2022.12 | 38.4 | — | |
| Bag of TricksBackbone=ResNet-32, Reference (Source of result)=Original paper2021.12 | — | 52.17 | |
| BBNBackbone=ResNet-32, Reference (Source of result)=Original paper2021.12 | — | 57.44 | |
| Class-balanced CEBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 61.23 | |
| Class-balanced fine-tuningBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 58.5 | |
| Cross-EntropyBackbone=ResNet-322021.12 | — | 61.54 | |
| Hybrid-PSCBackbone=ResNet-32, Reference (Source of result)=Original paper2021.12 | — | 55.03 | |
| L2RWBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 61.1 | |
| Meta-class-weight with cross-entropy lossBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 56.65 | |
| Meta-weight netBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 58.39 | |
| MetaSAug with cross-entropy lossBackbone=ResNet-32, Reference (Source of result)=Original paper2021.12 | — | 53.13 | |
| MixupBackbone=ResNet-32, Reference (Source of result)=Li et al. 20212021.12 | — | 60.46 | |
| RISDABackbone=ResNet-322021.12 | — | 49.84 |