Fine-grained Image Classification on Stanford Dogs (test)
97.3AccuracySR-GNN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SR-GNNBackbone=Xception, Attentional Refinement=true2022.09 | 97.3 | — | |
| CPMBackbone=GoogLeNet, Joint Learning=true2022.09 | 97.1 | — | |
| SR-GNNBackbone=Xception, Attentional Refinement=false2022.09 | 96.5 | — | |
| CAPBackbone=Xception2022.09 | 96.1 | — | |
| ViTintegrated_with_SAC=true2022.05 | 94.5 | — | |
| Parts Modelsintegrated_with_SAC=false2022.05 | 93.9 | — | |
| ViTintegrated_with_SAC=false2022.05 | 93.2 | — | |
| WS_DANintegrated_with_SAC=true2022.05 | 93.1 | — | |
| WARN2018.07 | 92.9 | — | |
| WARNBackbone=Wide ResNet-502022.09 | 92.9 | — | |
| CAMFBackbone=Swin Transformer, Vision Transformer=true2022.09 | 92.8 | — | |
| WSintegrated_with_SAC=true2022.05 | 92.5 | — | |
| TPSKGBackbone=ViT-B-16, Resolution=448 x 4482021.07 | 92.5 | — | |
| TrnFGBackbone=ViT-B-16, Text Description=true2022.09 | 92.3 | — | |
| WS_DANintegrated_with_SAC=false2022.05 | 92.2 | — | |
| DANBackbone=Inception-V32022.09 | 92.2 | — | |
| ViTBackbone=ViT-B-16, Vision Transformer=true2022.09 | 91.7 | — | |
| MMALintegrated_with_SAC=true2022.05 | 91.6 | — | |
| WSintegrated_with_SAC=false2022.05 | 91.4 | — | |
| ViTBackbone=ViT-B-16, Resolution=448 x 4482021.07 | 91.4 | — | |
| MMALintegrated_with_SAC=false2022.05 | 90.6 | — | |
| API-NetBackbone=ResNet-101, Extra Supervision=No2020.02 | 90.3 | — | |
| API-Netintegrated_with_SAC=false2022.05 | 90.3 | — | |
| API-NETBackbone=DenseNet-161, Resolution=512 x 5122021.07 | 90.3 | — | |
| APINBackbone=ResNet-1012022.09 | 90.3 | — | |
| WRN2018.07 | 89.6 | — | |
| API-NetBackbone=DenseNet-161, Extra Supervision=No2020.02 | 89.4 | — | |
| MRDMNBackbone=ResNet-502022.09 | 89.1 | — | |
| FCAN1-Stage=true, Sep. Init.=false2019.09 | 88.9 | — | |
| Cross-X (ResNet)1-Stage=true, Sep. Init.=false, Backbone=ResNet-502019.09 | 88.9 | — | |
| Cross-XBackbone=ResNet-50, Resolution=448 x 4482021.07 | 88.9 | — | |
| Cross-XBackbone=ResNet-502022.09 | 88.9 | — | |
| DTintegrated_with_SAC=true2022.05 | 88.8 | — | |
| API-NetBackbone=ResNet-50, Extra Supervision=No2020.02 | 88.3 | — | |
| Cross-X (SENet)1-Stage=true, Sep. Init.=false, Backbone=SENet-502019.09 | 88.2 | — | |
| ResNet-501-Stage=true, Sep. Init.=false2019.09 | 88.1 | — | |
| DTintegrated_with_SAC=false2022.05 | 88 | — | |
| DBBackbone=ResNet-50, Resolution=448 x 4482021.07 | 87.7 | — | |
| ViT-ResNet-50Backbone=ViT&ResNet-50, Resolution=448 x 4482021.07 | 87.7 | — | |
| ResNet-50integrated_with_SAC=true2022.05 | 87.4 | — | |
| RA-CNNHR (high-resolution)=true2018.07 | 87.3 | — | |
| RA-CNN1-Stage=false, Sep. Init.=true2019.09 | 87.3 | — | |
| RACNNBackbone=VGGNet-19, Extra Supervision=No2020.02 | 87.3 | — | |
| RA_CNNintegrated_with_SAC=false2022.05 | 87.3 | — | |
| RA-CNNInterpretability level=Part-level attention2021.11 | 87.3 | — | |
| SENet-501-Stage=true, Sep. Init.=false2019.09 | 87.1 | — | |
| Inception-V3integrated_with_SAC=true2022.05 | 86.8 | — | |
| MLA-CNNBackbone=VGG-19, Resolution=448 x 4482021.07 | 86.8 | — | |
| Deformable ProtoPNetInterpretability level=Part-level attention + prototypes, Deformation setting=No deformations (nd), Prototype configuration=3x3 (3p), Prototypes per class=10 (10pc)2021.11 | 86.5 | — | |
| Deformable ProtoPNetInterpretability level=Part-level attention + prototypes + deformations, Deformation setting=Deformable (3p), Prototype configuration=3x3 (3p), Prototypes per class=10 (10pc)2021.11 | 86.5 | — | |
| ResNet-50integrated_with_SAC=false2022.05 | 86.1 | — | |
| MSECBackbone=ResNet-50, Resolution=448 x 4482021.07 | 85.6 | — | |
| MAMCBackbone=ResNet-101, Extra Supervision=No2020.02 | 85.2 | — | |
| MAMCintegrated_with_SAC=false2022.05 | 85.2 | — | |
| Inception-V3integrated_with_SAC=false2022.05 | 85.1 | — | |
| FDLBackbone=DenseNet-161, Resolution=448 x 4482021.07 | 84.9 | — | |
| MAMC-CNN1-Stage=true, Sep. Init.=false2019.09 | 84.8 | — | |
| FCANBackbone=ResNet-50, Extra Supervision=No2020.02 | 84.2 | — | |
| FCANInterpretability level=Part-level attention2021.11 | 84.2 | — | |
| PCBackbone=DenseNet-161, Extra Supervision=No2020.02 | 83.8 | — | |
| PCintegrated_with_SAC=false2022.05 | 83.8 | — | |
| PC-CNNBackbone=DenseNet-161, Resolution=224 x 2242021.07 | 83.8 | — | |
| MaxEntBackbone=DenseNet-161, Extra Supervision=No2020.02 | 83.6 | — | |
| MaxEntBackbone=DenseNet-1612021.07 | 83.6 | — | |
| Base CNNBackbone=Xception2022.09 | 82.7 | — | |
| B-CNNBackbone=VGGNet-19, Extra Supervision=No2020.02 | 82.1 | — | |
| DivideMix + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 81.9 | — | |
| DVANBackbone=VGGNet-16, Extra Supervision=No2020.02 | 81.5 | — | |
| DVANBackbone=VGG-16, Resolution=224 x 2242021.07 | 81.5 | — | |
| DivideMixNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 79.39 | — | |
| MLC + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 79.22 | — | |
| EM-based FVEPre-training=iNaturalist 2017, Architecture=parts of [24]2020.07 | 79.2 | — | |
| DATLPre-training=iNaturalist 2017, Reference=[16]2020.07 | 79.1 | — | |
| Gradient-based FVEPre-training=iNaturalist 2017, Architecture=parts of [24]2020.07 | 79.1 | — | |
| MW-Net + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 78.52 | — | |
| Cui et al.Pre-training=iNaturalist 20172020.07 | 78.5 | — | |
| GAPPre-training=iNaturalist 2017, Architecture=parts of [24]2020.07 | 77.8 | — | |
| No Parts (baseline)Pre-training=iNaturalist 20172020.07 | 77.5 | — | |
| SYM + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 77.37 | — | |
| ProtoPNetInterpretability level=Part-level attention + prototypes2021.11 | 77.3 | — | |
| DivideMix + SNSCLNoise Type=Asymmetric, Noise Ratio=30%2023.03 | 77.19 | — | |
| Cross-Entropy + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 76.24 | — | |
| Conf. Penalty + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 76.01 | — | |
| MLC + SNSCLNoise Type=Asymmetric, Noise Ratio=30%2023.03 | 75.92 | — | |
| GCE + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 75.91 | — | |
| Label Smooth + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 75.84 | — | |
| MLCNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 75.84 | — | |
| DivideMixNoise Type=Asymmetric, Noise Ratio=30%2023.03 | 75.51 | — | |
| SYM + SNSCLNoise Type=Asymmetric, Noise Ratio=30%2023.03 | 74.74 | — | |
| Label SmoothNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 74.7 | — | |
| Conf. PenaltyNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 74.41 | — | |
| JoCoR + SNSCLNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 74.26 | — | |
| Cross-EntropyNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 74.24 | — | |
| MW-NetNoise Type=Asymmetric, Noise Ratio=10%2023.03 | 73.68 | — | |
| MaxEnt-CNN1-Stage=true, Sep. Init.=false2019.09 | 73.6 | — | |
| GBVBackbone=ResNet-34, tau parameter=best2026.04 | 73.2 | — | |
| GBVBackbone=ResNet-34, tau parameter=quick2026.04 | 73.16 | — | |
| MW-Net + SNSCLNoise Type=Asymmetric, Noise Ratio=30%2023.03 | 72.68 | — | |
| EntAugment (2024)Backbone=ResNet-342026.04 | 72.17 | — | |
| PDFRBackbone=AlexNet, Extra Supervision=No2020.02 | 72 | — |