Image Captioning on Image Captioning Target: GPT-5 Soft Criterion
56ASRCoTTA
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CoTTASurrogate=Ensemble2026.03 | 56 | 32 | |
| M-AttackSurrogate=Ensemble2026.03 | 32 | 14 | |
| FOA-AttackSurrogate=Ensemble2026.03 | 27 | 12.8 | |
| Agent-AttackSurrogate=Ensemble2026.03 | 10 | 4.9 | |
| AttackVLMSurrogate=B/162026.03 | 0 | 0 | |
| AttackVLMSurrogate=B/322026.03 | 0 | 0.2 | |
| AttackVLMSurrogate=Laion2026.03 | 0 | 0.2 | |
| AnyAttackSurrogate=Ensemble2026.03 | 0 | 0.2 |