Image Classification on FGVC Suite
93.32Average AccuracyGPS
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GPSArchitecture=ConvNeXt-B, Trainable Parameters (%)=0.902023.12 | 93.32 | — | — | — | — | — | — | |
| FullArchitecture=ConvNeXt-B, Trainable Parameters (%)=100.002023.12 | 93.04 | — | — | — | — | — | — | |
| GPSArchitecture=Swin-B, Trainable Parameters (%)=0.952023.12 | 92.56 | — | — | — | — | — | — | |
| SSFArchitecture=ConvNeXt-B, Trainable Parameters (%)=0.562023.12 | 92.48 | — | — | — | — | — | — | |
| FullArchitecture=Swin-B, Trainable Parameters (%)=100.002023.12 | 92.42 | — | — | — | — | — | — | |
| OursBackbone=ViT-Base/16, Tuned/Total (%)=0.75, Extra Params=true2026.06 | 91.6 | — | — | — | — | — | — | |
| SSFArchitecture=Swin-B, Trainable Parameters (%)=0.562023.12 | 91.54 | — | — | — | — | — | — | |
| LoR-VPBackbone=ViT-Base/16, Extra Params=true2026.06 | 91.22 | — | — | — | — | — | — | |
| HTABackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=true, Params. (M)=0.362024.10 | 91.1 | 90.5 | 99.6 | 90.5 | 89.3 | 85.4 | — | |
| RLRRBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=true, Params. (M)=0.472024.10 | 91 | 89.8 | 99.6 | 90.4 | 90 | 85.3 | — | |
| SSFBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.392024.10 | 90.7 | 89.5 | 99.6 | 89.2 | 89.6 | 85.7 | — | |
| ARCBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=true, Params. (M)=0.252024.10 | 90.7 | 89.3 | 99.7 | 89.5 | 89.1 | 85.7 | — | |
| HTABackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.362024.10 | 90.6 | 88.8 | 99.5 | 87.9 | 92.2 | 84.4 | — | |
| RLRRBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.472024.10 | 90.4 | 89.3 | 99.5 | 87 | 92 | 84.7 | — | |
| GenFTParam. (M)=0.272025.05 | 90.38 | 89.3 | 99.3 | 86.2 | 91.5 | 85.6 | — | |
| ARCBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.252024.10 | 90.1 | 88.5 | 99.3 | 85.7 | 91.9 | 85.3 | — | |
| SPTBackbone=ViT-Base/16, Extra Params=true2026.06 | 90.1 | — | — | — | — | — | — | |
| SA2VPParam. (M)=0.682025.05 | 90.08 | 89.1 | 99.3 | 84.1 | 92.1 | 85.8 | — | |
| SA2VPBackbone=ViT-Base/16, Tuned/Total (%)=0.81, Extra Params=true2026.06 | 90.08 | — | — | — | — | — | — | |
| SPT-LoRAParam. (M)=0.432025.05 | 90.04 | 88.6 | 99.5 | 87.3 | 91.4 | 83.4 | — | |
| S2FTParams (%)=0.162026.05 | 89.6 | 88.4 | 99.3 | 84.5 | 91.4 | 84.6 | — | |
| VAPTParam. (M)=0.192025.05 | 89.58 | 89.7 | 99.1 | 82.8 | 91.7 | 84.6 | — | |
| LORABackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.442024.10 | 89.5 | 88.3 | 99.2 | 83.2 | 91 | 85.6 | — | |
| LoRABackbone=ViT-Base/16, Extra Params=true2026.06 | 89.46 | — | — | — | — | — | — | |
| DA-VPTBackbone=ViT-Base/16, Extra Params=true2026.06 | 89.32 | — | — | — | — | — | — | |
| VFPTParam. (M)=0.452025.05 | 89.24 | 88.7 | 99.1 | 83.6 | 90.4 | 84.5 | — | |
| VFPTBackbone=ViT-Base/16, Tuned/Total (%)=0.66, Extra Params=true2026.06 | 89.24 | — | — | — | — | — | — | |
| E2VPTParam. (M)=0.282025.05 | 89.22 | 89.1 | 99.1 | 82.8 | 90.5 | 84.6 | — | |
| E2VPTBackbone=ViT-Base/16, Tuned/Total (%)=0.39, Extra Params=true2026.06 | 89.22 | — | — | — | — | — | — | |
| ARCattBackbone=ViT-Base/16, Extra Params=true2026.06 | 89.12 | — | — | — | — | — | — | |
| VPT-DBackbone=ViT-Base/16, Tuned/Total (%)=0.73, Extra Params=true2026.06 | 89.11 | — | — | — | — | — | — | |
| VPT-DeepBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.852024.10 | 89.1 | 88.5 | 99 | 83.6 | 90.2 | 84.2 | — | |
| S2FTParams (%)=0.082026.05 | 89.1 | 87.9 | 99.1 | 83.6 | 90.6 | 84.3 | — | |
| VPTParam. (M)=0.562025.05 | 89.1 | 88.5 | 99 | 83.6 | 90.2 | 84.2 | — | |
| HydraLoRAParams (%)=0.472026.05 | 89 | 86.9 | 99.1 | 89.1 | 88.2 | 82.1 | — | |
| Full FTParam. (M)=85.882025.05 | 88.54 | 87.3 | 98.8 | 84.5 | 89.4 | 82.7 | — | |
| FullBackbone=ViT-Base/16, Tuned/Total (%)=100.002026.06 | 88.54 | — | — | — | — | — | — | |
| Full fine-tuningBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=85.82024.10 | 88.5 | 87.3 | 98.8 | 84.5 | 89.4 | 82.7 | — | |
| FullParams (%)=100.002026.05 | 88.5 | 87.3 | 98.8 | 84.5 | 89.4 | 82.7 | — | |
| LoRA+Params (%)=0.342026.05 | 88.5 | 86.5 | 99.2 | 88.7 | 87.4 | 80.8 | — | |
| AdaptFormerParam. (M)=0.162025.05 | 88.48 | 88.4 | 99.2 | 81.9 | 88.2 | 84.7 | — | |
| BiasBackbone=ViT-Base/16, Tuned/Total (%)=0.802026.06 | 88.41 | — | — | — | — | — | — | |
| BiasBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.282024.10 | 88.4 | 88.4 | 98.8 | 79.4 | 91.2 | 84.2 | — | |
| BiasParams (%)=0.332026.05 | 88.4 | 88.4 | 98.8 | 79.4 | 91.2 | 84.2 | — | |
| BitFitParam. (M)=0.102025.05 | 88.02 | 87.7 | 99.2 | 81.5 | 86.5 | 85.2 | — | |
| LinearArchitecture=ConvNeXt-B, Trainable Parameters (%)=0.282023.12 | 88 | — | — | — | — | — | — | |
| LinearArchitecture=Swin-B, Trainable Parameters (%)=0.282023.12 | 87.9 | — | — | — | — | — | — | |
| FourierFTParams (%)=0.162026.05 | 87.9 | 85.5 | 99.1 | 83.7 | 89 | 82.4 | — | |
| LoRAParams (%)=0.342026.05 | 87.5 | 84.9 | 98.1 | 87.4 | 88.1 | 79 | — | |
| LoRA-FAParams (%)=0.172026.05 | 86.4 | 85.8 | 99.1 | 80.1 | 88.1 | 79.1 | — | |
| AdapterBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.412024.10 | 85.7 | 87.1 | 98.5 | 68.6 | 89.8 | 84.3 | — | |
| AdapterParam. (M)=0.162025.05 | 85.66 | 87.1 | 98.5 | 68.6 | 89.8 | 84.3 | — | |
| AdapterParams (%)=0.482026.05 | 85.6 | 87.1 | 98.5 | 68.6 | 89.8 | 84.3 | — | |
| AdapterBackbone=ViT-Base/16, Tuned/Total (%)=1.02, Extra Params=true2026.06 | 85.46 | — | — | — | — | — | — | |
| LoRAParam. (M)=0.432025.05 | 84.78 | 85.6 | 98.9 | 72 | 87.6 | 79.8 | — | |
| VPT-SBackbone=ViT-Base/16, Tuned/Total (%)=0.16, Extra Params=true2026.06 | 84.62 | — | — | — | — | — | — | |
| VPT-ShallowBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.252024.10 | 84.6 | 86.7 | 98.4 | 68.7 | 90.7 | 78.8 | — | |
| VPT-DEEPSSL=MoCo v3, TOTAL PARAMS=1.02x2023.06 | 83.12 | 82.67 | 94.41 | 79.18 | 83.33 | 75.99 | — | |
| Gated Prompt TuningSSL=MoCo v3, TOTAL PARAMS=1.02x2023.06 | 83 | 82.86 | 93.71 | 79.02 | 83.37 | 76.02 | — | |
| Partial-1Backbone=ViT-Base/16, Tuned/Total (%)=8.342026.06 | 82.63 | — | — | — | — | — | — | |
| A-SelecT2026.03 | 82.5 | 78.6 | 90.6 | 86.1 | 83.5 | 78.4 | 77.5 | |
| MLP-3Backbone=ViT-Base/16, Tuned/Total (%)=1.44, Extra Params=true2026.06 | 79.8 | — | — | — | — | — | — | |
| LinearParam. (M)=0.002025.05 | 79.32 | 85.3 | 97.9 | 51.3 | 86.2 | 75.9 | — | |
| LinearBackbone=ViT-Base/16, Tuned/Total (%)=0.082026.06 | 79.32 | — | — | — | — | — | — | |
| Linear probingBackbone=ViT-B/16, Pre-trained=ImageNet-21K, AugReg=false, Params. (M)=0.182024.10 | 79.3 | 85.3 | 97.9 | 51.3 | 86.2 | 75.9 | — | |
| LinearParams (%)=0.212026.05 | 79.3 | 85.3 | 97.9 | 51.3 | 86.2 | 75.9 | — | |
| VPT-SHALLOWSSL=MoCo v3, TOTAL PARAMS=1.02x2023.06 | 79.26 | 79.05 | 90.47 | 71.91 | 81.97 | 72.92 | — | |
| MAGE2026.03 | 78.4 | 77.7 | 86.8 | 73.4 | 88.2 | 76.7 | 67.6 | |
| SidetuneBackbone=ViT-Base/16, Tuned/Total (%)=10.082026.06 | 78.35 | — | — | — | — | — | — | |
| SDXL2026.03 | 78.3 | 71.9 | 87.5 | 83.2 | 82.1 | 71.6 | 73.7 | |
| DifFeed2026.03 | 78.2 | 71 | 90.1 | 82.7 | 81.3 | 70.7 | 73.2 | |
| Gated Prompt TuningSSL=MAE, TOTAL PARAMS=1.02x2023.06 | 73.39 | 70.56 | 78.55 | 71.7 | 78.9 | 67.26 | — | |
| VPT-DEEPSSL=MAE, TOTAL PARAMS=1.02x2023.06 | 72.02 | 68.33 | 80.05 | 67.67 | 78.83 | 65.22 | — | |
| ResNet-50backbone=frozen2026.03 | 69.3 | 64 | 82.5 | 68.8 | 82.3 | 54.2 | 63.8 | |
| SwAV2026.03 | 66.7 | 63.8 | 91.8 | 57.6 | 74.9 | 54.1 | 58.2 | |
| VPT-SHALLOWSSL=MAE, TOTAL PARAMS=1.02x2023.06 | 57.84 | 42.15 | 69.15 | 43.38 | 77.07 | 57.43 | — | |
| SimCLR2026.03 | 52.9 | 43.9 | 86.2 | 40.9 | 64.2 | 35.9 | 46.4 | |
| GD2026.03 | 45.2 | 32.8 | 78.4 | 24.5 | 58.3 | 20.1 | 57.2 | |
| APArchitecture=ResNet-1012026.04 | — | 69.42 | 85.14 | 59.06 | 87.79 | 58.31 | — | |
| APArchitecture=ViT-L/162026.04 | — | 86.74 | 99.7 | 69.41 | 90.83 | 79.83 | — | |
| Full-TuneArchitecture=ResNet-1012026.04 | — | 88.91 | 99.98 | 87.76 | 90.13 | 84.45 | — | |
| Full-TuneArchitecture=ViT-L/162026.04 | — | 89.79 | 99.91 | 89.42 | 93.31 | 84.75 | — | |
| Linear-ProbeArchitecture=ResNet-1012026.04 | — | 63.76 | 82.01 | 49.62 | 86.63 | 52.09 | — | |
| Linear-ProbeArchitecture=ViT-L/162026.04 | — | 84.69 | 99.4 | 65.24 | 86.11 | 75.71 | — | |
| Norm-TuneArchitecture=ResNet-1012026.04 | — | 66.39 | 66.5 | 67.64 | 87.59 | 56.72 | — | |
| Norm-TuneArchitecture=ViT-L/162026.04 | — | 85.9 | 99.35 | 75.61 | 89.76 | 78.78 | — | |
| VPArchitecture=ResNet-1012026.04 | — | 65.72 | 78.5 | 51.04 | 86.91 | 54.23 | — | |
| VPArchitecture=ViT-L/162026.04 | — | 85.24 | 99.32 | 67.64 | 87.02 | 76.2 | — |