Image Classification on iNaturalist 2017 (test)
71.7Top-1 AccuracyTransFG
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| TransFGBackbone=ViT-B_162021.03 | 71.7 | — | — | — | — | — | — | — | |
| ViTBackbone=ViT-B_162021.03 | 68.7 | — | — | — | — | — | — | — | |
| TASNBackbone=ResNet1012021.03 | 68.2 | — | — | — | — | — | — | — | |
| IncResNetV2Backbone=IncResNetV22021.03 | 67.3 | — | — | — | — | — | — | — | |
| ResNet152Backbone=ResNet1522021.03 | 59 | — | — | — | — | — | — | — | |
| Explicit EnsembleEnsemble size=4, Random seeds=32024.05 | 49.6 | — | — | — | 44.6 | 19.9 | 0.716 | 16.5 | |
| LoRA-EnsembleEnsemble size=4, Random seeds=32024.05 | 49.3 | — | — | — | 44.1 | 4.5 | 0.61 | 16 | |
| Single Net w/ LoRAEnsemble size=1, Random seeds=32024.05 | 47.7 | — | — | — | 43.1 | 9.6 | 0.662 | 16.6 | |
| MC DropoutEnsemble size=1, Random seeds=32024.05 | 47.5 | — | — | — | 40.3 | 20.6 | 0.895 | 17.2 | |
| Spectral Co-distillationBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 45.87 | — | — | — | — | — | — | — | |
| FedRoDBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 44.54 | — | — | — | — | — | — | — | |
| Single NetworkEnsemble size=1, Random seeds=32024.05 | 42.6 | — | — | — | 37.8 | 29.3 | 1.054 | 20.7 | |
| FedBABUBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 42.36 | — | — | — | — | — | — | — | |
| FedRepBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 42.07 | — | — | — | — | — | — | — | |
| DittoBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 41.88 | — | — | — | — | — | — | — | |
| Spectral Co-distillationBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 41.75 | — | — | — | — | — | — | — | |
| FedProxBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 41.58 | — | — | — | — | — | — | — | |
| FedDynBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 40.99 | — | — | — | — | — | — | — | |
| FedRoDBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 40.16 | — | — | — | — | — | — | — | |
| FedRepBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 39.81 | — | — | — | — | — | — | — | |
| FedProxBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 39.46 | — | — | — | — | — | — | — | |
| FedDynBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 39.35 | — | — | — | — | — | — | — | |
| DittoBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 39.33 | — | — | — | — | — | — | — | |
| FedBABUBackbone=ResNet-50, Number of clients (N)=20, Data heterogeneity (alpha)=0.1, Trials=32024.01 | 39.23 | — | — | — | — | — | — | — | |
| BBNBackbone=ResNet-50, Training schedule=1x2019.12 | — | 36.61 | — | — | — | — | — | — | |
| BBNBackbone=ResNet-50, Training schedule=2x2019.12 | — | 34.25 | — | — | — | — | — | — | |
| BBNquoted from original paper=true2021.03 | — | 36.61 | — | — | — | — | — | — | |
| CB CEBackbone=ResNet-50, Pre-training=ImageNet2020.03 | — | 42.59 | 25.92 | 20.6 | — | — | — | — | |
| CB focalBackbone=ResNet-50, Pre-training=ImageNet2020.03 | — | 41.92 | — | 20.92 | — | — | — | — | |
| CB-FocalBackbone=ResNet-502019.12 | — | 41.92 | — | — | — | — | — | — | |
| CEBackbone=ResNet-502019.12 | — | 45.38 | — | — | — | — | — | — | |
| CEBackbone=ResNet-50, Pre-training=ImageNet2020.03 | — | 43.49 | 26.6 | 21 | — | — | — | — | |
| CEloss function=Cross-entropy2021.03 | — | 43.21 | — | — | — | — | — | — | |
| CE-DRSBackbone=ResNet-502019.12 | — | 40.12 | — | — | — | — | — | — | |
| CE-DRWBackbone=ResNet-502019.12 | — | 40.48 | — | — | — | — | — | — | |
| Class-balanced CE2021.03 | — | 42.02 | — | — | — | — | — | — | |
| Class-balanced focalquoted from original paper=true2021.03 | — | 41.92 | — | — | — | — | — | — | |
| LDAM2021.03 | — | 39.15 | — | — | — | — | — | — | |
| LDAM-DRWBackbone=ResNet-50, Training schedule=1x2019.12 | — | 39.49 | — | — | — | — | — | — | |
| LDAM-DRWBackbone=ResNet-50, Training schedule=2x2019.12 | — | 38.19 | — | — | — | — | — | — | |
| LDAM-DRW2021.03 | — | 37.84 | — | — | — | — | — | — | |
| Meta-class-weightreported in=[18]2021.03 | — | 40.62 | — | — | — | — | — | — | |
| MetaSAug2021.03 | — | 36.72 | — | — | — | — | — | — | |
| Ours, CEBackbone=ResNet-50, Pre-training=ImageNet2020.03 | — | 40.62 | 23.7 | 18.4 | — | — | — | — |