Adversarial Defense on ImageNet-1K 1.0 (test)
78.6Classification AccuracyPlatonic Defense
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Platonic DefenseEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | 78.6 | 63.68 | 0.76 | 36.32 | |
| No DefenseEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | 78.3 | 0.44 | — | 99.56 | |
| DecompEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | 78.22 | 0.84 | 0.46 | 99.16 | |
| ZIPEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | 67.99 | 55.42 | — | 44.58 | |
| DeDeEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | 46.78 | 4.66 | 0.53 | 95.34 | |
| BeatrixEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | — | — | 0.39 | — | |
| DetectCLIPEvaluation protocol=linear probing, Attack=Average (Adversarial Attacks)2026.06 | — | — | 0.79 | — |