Image Classification on iNaturalist (test)
81.2AccuracyFinetune
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Finetunelabeled_data_ratio=100%, backbone=ViT-Base2022.08 | 81.2 | — | |
| TResNet-LPre-training Dataset=ImageNet-21K-P2021.04 | 74.8 | — | |
| TResNet-MPre-training Dataset=ImageNet-21K-P2021.04 | 72.7 | — | |
| TResNet-LPre-training Dataset=ImageNet-1K2021.04 | 72.4 | — | |
| OFA595Pre-training Dataset=ImageNet-21K-P2021.04 | 71.5 | — | |
| ResNet50Pre-training Dataset=ImageNet-21K-P2021.04 | 71.4 | — | |
| TResNet-MPre-training Dataset=ImageNet-1K2021.04 | 70.1 | — | |
| OFA595Pre-training Dataset=ImageNet-1K2021.04 | 69 | — | |
| Semi-ViTlabeled_data_ratio=10%, backbone=ViT-Base2022.08 | 67.7 | — | |
| ResNet50Pre-training Dataset=ImageNet-1K2021.04 | 66.8 | — | |
| MobileNetV3Pre-training Dataset=ImageNet-21K-P2021.04 | 65 | — | |
| MobileNetV3Pre-training Dataset=ImageNet-1K2021.04 | 62.4 | — | |
| Finetunelabeled_data_ratio=10%, backbone=ViT-Base2022.08 | 57.3 | — | |
| PEP-FedPT2025.10 | 54.16 | — | |
| SGPT2025.10 | 53.81 | — | |
| Fed-VPT-D2025.10 | 53.2 | — | |
| SEERArch.=RG256, Pre-training source=uncurated data, Evaluation Protocol=linear evaluation2021.03 | 50.8 | — | |
| SwAVArch.=RN50, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 48.6 | — | |
| Fed-VPT2025.10 | 48.05 | — | |
| SwAVArch.=RG128, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 47.5 | — | |
| SPAM (Neural, Order 3)Interaction Order=3, Model Architecture=Neural2022.05 | 47.22 | — | |
| SupervisedArch.=RG128, Pre-training source=ImageNet, Evaluation Protocol=linear evaluation2021.03 | 47.2 | — | |
| SEERArch.=RG128, Pre-training source=uncurated data, Evaluation Protocol=linear evaluation2021.03 | 47.2 | — | |
| SPAM (Neural, Order 2)Interaction Order=2, Model Architecture=Neural2022.05 | 46.89 | — | |
| SPAM (Linear, Order 3)Interaction Order=3, Model Architecture=Linear2022.05 | 46.84 | — | |
| SPAM (Linear, Order 2)Interaction Order=2, Model Architecture=Linear2022.05 | 46.05 | — | |
| DNNModel Architecture=Neural2022.05 | 45.84 | — | |
| Head2025.10 | 45.27 | — | |
| Linear (Order 2)Interaction Order=2, Model Architecture=Linear2022.05 | 42.92 | — | |
| NAMInteraction Order=1, Model Architecture=Neural2022.05 | 41.94 | — | |
| P-PT2025.10 | 41.2 | — | |
| Linear (Order 1)Interaction Order=1, Model Architecture=Linear2022.05 | 39.32 | — | |
| FedPR2025.10 | 36.03 | — | |
| Semi-ViTlabeled_data_ratio=1%, backbone=ViT-Base2022.08 | 32.3 | — | |
| Finetunelabeled_data_ratio=1%, backbone=ViT-Base2022.08 | 19.6 | — | |
| AdaptiveBackbone=ResNet-50, Epochs=902020.07 | — | 35.42 | |
| AdaptiveBackbone=ResNet-152, Epochs=902020.07 | — | 31.18 | |
| AdaptiveBackbone=ResNet-152, Epochs=2002020.07 | — | 29.46 | |
| EqualisedBackbone=ResNet-50, Epochs=902020.07 | — | 38.37 | |
| EqualisedBackbone=ResNet-152, Epochs=902020.07 | — | 35.86 | |
| EqualisedBackbone=ResNet-152, Epochs=2002020.07 | — | 34.53 | |
| ERMBackbone=ResNet-50, Epochs=902020.07 | — | 38.66 | |
| ERMBackbone=ResNet-152, Epochs=902020.07 | — | 35.88 | |
| ERMBackbone=ResNet-152, Epochs=2002020.07 | — | 34.38 | |
| Logit adjustment loss (τ = 1)Backbone=ResNet-50, Epochs=902020.07 | — | 33.64 | |
| Logit adjustment loss (τ = 1)Backbone=ResNet-152, Epochs=902020.07 | — | 31.15 | |
| Logit adjustment loss (τ = 1)Backbone=ResNet-152, Epochs=2002020.07 | — | 30.12 | |
| Logit adjustment plus adaptive loss (τ = 1)Backbone=ResNet-50, Epochs=902020.07 | — | 31.56 | |
| Logit adjustment plus adaptive loss (τ = 1)Backbone=ResNet-152, Epochs=902020.07 | — | 29.22 | |
| Logit adjustment plus adaptive loss (τ = 1)Backbone=ResNet-152, Epochs=2002020.07 | — | 28.02 | |
| Logit adjustment post-hoc (τ = 1)Backbone=ResNet-50, Epochs=902020.07 | — | 33.98 | |
| Logit adjustment post-hoc (τ = 1)Backbone=ResNet-152, Epochs=902020.07 | — | 31.46 | |
| Logit adjustment post-hoc (τ = 1)Backbone=ResNet-152, Epochs=2002020.07 | — | 30.15 | |
| Logit adjustment post-hoc (τ = τ*)Backbone=ResNet-50, Epochs=902020.07 | — | 33.8 | |
| Logit adjustment post-hoc (τ = τ*)Backbone=ResNet-152, Epochs=902020.07 | — | 31.08 | |
| Logit adjustment post-hoc (τ = τ*)Backbone=ResNet-152, Epochs=2002020.07 | — | 29.74 | |
| Weight normalisation (τ = 1)Backbone=ResNet-50, Epochs=902020.07 | — | 48.05 | |
| Weight normalisation (τ = 1)Backbone=ResNet-152, Epochs=902020.07 | — | 45.17 | |
| Weight normalisation (τ = 1)Backbone=ResNet-152, Epochs=2002020.07 | — | 45.33 | |
| Weight normalisation (τ = τ*)Backbone=ResNet-50, Epochs=902020.07 | — | 34.1 | |
| Weight normalisation (τ = τ*)Backbone=ResNet-152, Epochs=902020.07 | — | 31.85 | |
| Weight normalisation (τ = τ*)Backbone=ResNet-152, Epochs=2002020.07 | — | 30.34 |