Image Classification on Image classification tasks
60.5FCSSmoothed FARE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Smoothed FAREModel=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025), Defense=FARE + Feature-Space Smoothing (FS-GSB)2026.01 | 60.5 | 48.4 | 19.2 | |
| Smoothed TeCoAModel=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025), Defense=TeCoA + Feature-Space Smoothing (FS-GSB)2026.01 | 57.6 | 32 | 9.6 | |
| Smoothed org.Model=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025), Defense=Feature-Space Smoothing (FS-GSB)2026.01 | 51.2 | 35.2 | 24.2 | |
| TeCoAModel=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025)2026.01 | 45.4 | 12 | 29.6 | |
| FAREModel=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025)2026.01 | 40.8 | 10.4 | 64.2 | |
| Org.Model=LLaVA-1.5-7B, Attack bound (epsilon)=32/255, Evaluation attack=FOA (Jia et al., 2025)2026.01 | 37 | 3.6 | 78.6 |