Adversarial Example Rectification on CIFAR-10 (test)
99.2Acc (FGSM L∞)Proposed method (BIM)
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Proposed method (BIM)Approach=Re-attack-based rectification, Re-attack method=BIM2026.01 | 99.2 | 99.7 | 100 | 100 | |
| Proposed method (DF)Approach=Re-attack-based rectification, Re-attack method=DF2026.01 | 99.1 | 99.5 | 99.7 | 99.8 | |
| Proposed method (FGSM)Approach=Re-attack-based rectification, Re-attack method=FGSM2026.01 | 99 | 99.7 | 100 | 100 | |
| Previous method (Kao et al.)Approach=XAI-based rectification2026.01 | 58.1 | 61.6 | 72.9 | 93.6 | |
| Denoising autoencoderApproach=Input transformation2026.01 | 45.5 | 44.6 | 61.7 | 73.1 | |
| Full-image Gaussian blurApproach=Input transformation2026.01 | 27.9 | 27.1 | 32.2 | 27.7 | |
| Random pixel replacementApproach=Input transformation2026.01 | 19 | 12 | 18 | 25 | |
| JPEG compressionApproach=Input transformation2026.01 | 9.3 | 0 | 5.1 | 40.4 |