Image Classification on Stanford Dogs (test)
92.4Top-1 AccRAMS-Trans
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RAMS-TransBackbone=ViT-B_162021.07 | 92.4 | — | |
| TransFGBackbone=ViT-B/162021.03 | 92.3 | — | |
| WS-DANCropping Strategy=Attention Cropping2019.01 | 92.2 | — | |
| ViTBackbone=ViT-B_162021.07 | 92.2 | — | |
| ViTBackbone=ViT-B/162021.03 | 91.7 | — | |
| FFVTBackbone=ViT-B_162021.07 | 91.5 | — | |
| TransFGBackbone=ViT-B_162021.07 | 90.6 | — | |
| API-NetBackbone=ResNet-1012021.07 | 90.3 | — | |
| API-NetBackbone=ResNet-1012021.03 | 90.3 | — | |
| API-NetBackbone=DenseNet1612021.07 | 90.3 | — | |
| ViTBackbone=ViT-B_162021.07 | 90.2 | — | |
| TransFG & PSMBackbone=ViT-B_162021.07 | 90 | — | |
| REGSLBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 89.58 | — | |
| l2-SPBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 88.95 | — | |
| FedLAWD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 88.95 | — | |
| Inception-V32019.01 | 88.9 | — | |
| Cross-XBackbone=ResNet-502021.07 | 88.9 | — | |
| Cross-XBackbone=ResNet-502021.03 | 88.9 | — | |
| Cross-XBackbone=ResNet502021.07 | 88.9 | — | |
| FedAvgD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 88.85 | — | |
| MoCo V2 Anisotropictraining_mode=SSL, texture_suppression=anisotropic2020.11 | 88.81 | — | |
| SEFBackbone=ResNet-502021.07 | 88.8 | — | |
| SEFBackbone=ResNet502021.07 | 88.8 | — | |
| FedLWSD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 88.75 | — | |
| FedAdpD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 88.68 | — | |
| FedHAWD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 88.13 | — | |
| FedHyperD_alpha (Heterogeneity Level)=1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 87.86 | — | |
| Elastic TrainerBackbone=ViT2025.10 | 87.79 | — | |
| DBBackbone=ResNet502021.07 | 87.7 | — | |
| RA-CNNCropping Strategy=Random Cropping2019.01 | 87.3 | — | |
| RA-CNNBackbone=VGG-192021.07 | 87.3 | — | |
| RA-CNNBackbone=VGG192021.07 | 87.3 | — | |
| RA-CNNInterpretability=Part-level attention2021.11 | 87.3 | — | |
| MoCo V2training_mode=SSL2020.11 | 87.13 | — | |
| FedHAWD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 86.76 | — | |
| Deformable ProtoPNet (nd)Interpretability=Part-level attn. + learned prototypes2021.11 | 86.5 | — | |
| Deformable ProtoPNetInterpretability=Part-level attn. + learned prototypes + deformations2021.11 | 86.5 | — | |
| Deformable ProtoPNet (without deformations)Backbone=ResNet-1522021.11 | 86.5 | — | |
| Deformable ProtoPNetBackbone=ResNet-1522021.11 | 86.5 | — | |
| l2-PGMBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 86.48 | — | |
| Last-K LayersBackbone=ViT2025.10 | 86.45 | — | |
| Full TrainingBackbone=ViT2025.10 | 86.42 | — | |
| Transfer LearningBackbone=ViT2025.10 | 86.16 | — | |
| FedLAWD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 85.9 | — | |
| ResNet-1012019.01 | 85.8 | — | |
| Supervised (Reproduced)training_mode=supervised2020.11 | 85.35 | — | |
| MAMCCropping Strategy=Random Cropping2019.01 | 85.2 | — | |
| BaselineBackbone=ResNet-1522021.11 | 85.2 | — | |
| AdaBetBackbone=ViT2025.10 | 85.12 | — | |
| MaxEntBackbone=DenseNet-1612021.07 | 84.9 | — | |
| FDLBackbone=DenseNet-1612021.07 | 84.9 | — | |
| FDLBackbone=DenseNet-1612021.03 | 84.9 | — | |
| FDLBackbone=DenseNet1612021.07 | 84.9 | — | |
| SaSPAAugmentation Method=SaSPA2024.06 | 84.3 | — | |
| FCANCropping Strategy=Random Cropping2019.01 | 84.2 | — | |
| FCANInterpretability=Part-level attention2021.11 | 84.2 | — | |
| LSBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 84.14 | — | |
| BaselineBackbone=Densenet1612021.11 | 84.1 | — | |
| DeepTaxonCategory=w/ Retrieval2026.04 | 84.1 | — | |
| Inception V3Million Mult-Adds=5000, Million Parameters=23.22017.04 | 84 | — | |
| CAL-AugAugmentation Method=CAL-Aug2024.06 | 83.9 | — | |
| FedAdpD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 83.84 | — | |
| PCCropping Strategy=Random Cropping2019.01 | 83.8 | — | |
| Deformable ProtoPNet (without deformations)Backbone=Densenet1612021.11 | 83.7 | — | |
| Deformable ProtoPNetBackbone=Densenet1612021.11 | 83.7 | — | |
| MaxEntBackbone=DenseNet-1612021.03 | 83.6 | — | |
| MaxEntBackbone=DenseNet1612021.07 | 83.6 | — | |
| Real GuidanceAugmentation Method=Real Guidance2024.06 | 83.5 | — | |
| MobileNetwidth_multiplier=1, resolution=224, Million Mult-Adds=569, Million Parameters=3.32017.04 | 83.3 | — | |
| FedAvgD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 83.26 | — | |
| FedLWSD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 83.16 | — | |
| l2-NormBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 83 | — | |
| FedHyperD_alpha (Heterogeneity Level)=0.1, Backbone=frozen ImageNet-21k-pretrained ViT + FC, Communication Rounds (T)=100, Local Epochs=10, Batch Size=64, Learning Rate (eta)=5.0 x 10^-5, Optimizer=Adam2026.05 | 82.98 | — | |
| Fine-tuningBackbone=ResNet-101, Pre-trained=ImageNet (ILSVRC-2012), Evaluation Protocol=Fine-tuning2021.11 | 82.89 | — | |
| ALIAAugmentation Method=ALIA2024.06 | 82.8 | — | |
| PruneTrainBackbone=ViT2025.10 | 82.6 | — | |
| MobileNetwidth_multiplier=0.75, resolution=224, Million Mult-Adds=325, Million Parameters=1.92017.04 | 81.9 | — | |
| MobileNetwidth_multiplier=1, resolution=192, Million Mult-Adds=418, Million Parameters=3.32017.04 | 81.9 | — | |
| AdaBetBackbone=MobileNetV22025.10 | 81.76 | — | |
| LiT-L16L + RECO# params (M)=652, zero-shot=true2023.06 | 81.3 | — | |
| MobileNetwidth_multiplier=0.75, resolution=192, Million Mult-Adds=239, Million Parameters=1.92017.04 | 80.5 | — | |
| Full TrainingBackbone=MobileNetV22025.10 | 80.47 | — | |
| Fisher InformationBackbone=MobileNetV22025.10 | 80.47 | — | |
| Transfer LearningBackbone=MobileNetV22025.10 | 80.4 | — | |
| Last-K LayersBackbone=MobileNetV22025.10 | 79.62 | — | |
| Deformable ProtoPNetBackbone=VGG-192021.11 | 77.9 | — | |
| ProtoPNetInterpretability=Part-level attn. + learned prototypes2021.11 | 77.3 | — | |
| BaselineBackbone=VGG-192021.11 | 77.3 | — | |
| ProtoPNetBackbone=Densenet1612021.11 | 77.3 | — | |
| Elastic TrainerBackbone=MobileNetV22025.10 | 77.02 | — | |
| VGG-192019.01 | 76.7 | — | |
| ProtoPNetBackbone=ResNet-1522021.11 | 76.2 | — | |
| LiT-L16L# params (M)=638, zero-shot=true2023.06 | 75.7 | — | |
| AdaBetBackbone=ResNet502025.10 | 75.52 | — | |
| Deformable ProtoPNet (without deformations)Backbone=VGG-192021.11 | 74.8 | — | |
| CLIP-L/14 + RECO# params (M)=435, zero-shot=true2023.06 | 73.9 | — | |
| Fisher InformationBackbone=ResNet502025.10 | 73.76 | — | |
| ProtoPNetBackbone=VGG-192021.11 | 73.6 | — | |
| PruneTrainBackbone=MobileNetV22025.10 | 73.31 | — | |
| Last-K LayersBackbone=ResNet502025.10 | 72.8 | — |