Image Classification on Flowers-102 (Robustness Evaluation)
87.87Accuracy (Clean)AdvMaPLe
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| AdvMaPLeShots=16, Training Epochs=10, Training Attack (PGD)=2-step (1/255), Evaluation Attack (PGD-100)=100-step (1/255)2026.05 | 87.87 | 58.7 | — | |
| AdvVLPShots=16, Training Epochs=10, Training Attack (PGD)=2-step (1/255), Evaluation Attack (PGD-100)=100-step (1/255)2026.05 | 87.7 | 58.7 | — | |
| FAPShots=16, Training Epochs=10, Training Attack (PGD)=2-step (1/255), Evaluation Attack (PGD-100)=100-step (1/255)2026.05 | 86.27 | 61.47 | — | |
| CLBPShots=16, Training Epochs=10, Training Attack (PGD)=2-step (1/255), Evaluation Attack (PGD-100)=100-step (1/255)2026.05 | 84.13 | 75.48 | — | |
| ConvNeXt-B + ConvStemArchitecture=CN-B + CvSt, Resolution=224x224, Perturbation radius (epsilon)=8/255, Fine-tuning protocol=fine-tuning from eps=4/255 for 25 epochs2023.03 | 83.98 | 54.77 | — | |
| XCIT-SArchitecture=XCIT-S, Resolution=224x224, Perturbation radius (epsilon)=8/255, Fine-tuning protocol=fine-tuning from eps=8/2552023.03 | 82.86 | 47.91 | — | |
| ConvNeXt-T + ConvStemArchitecture=CN-T + CvSt, Resolution=224x224, Perturbation radius (epsilon)=8/255, Fine-tuning protocol=fine-tuning from eps=4/255 for 25 epochs2023.03 | 80.42 | 50.89 | — | |
| AdvVPShots=16, Training Epochs=10, Training Attack (PGD)=2-step (1/255), Evaluation Attack (PGD-100)=100-step (1/255)2026.05 | 56.17 | 22.03 | — |