Image Captioning on COCO (CIDEr)
126.49CIDErCLIP
Evaluation Results
| Method | Links | |
|---|---|---|
| CLIPPerturbation Budget=clean, Attack=APGD2025.02 | 126.49 | |
| DSCA2026.04 | 123.8 | |
| AdvXLCLIP-HAttack Type=Clean, Perturbation budget (epsilon)=02026.06 | 122.66 | |
| DualEdit2026.04 | 121.4 | |
| CLIPAttack Type=Clean, Perturbation budget (epsilon)=02026.06 | 121.3 | |
| LiveEdit2026.04 | 120.6 | |
| AdvXLCLIP-LAttack Type=Clean, Perturbation budget (epsilon)=02026.06 | 120.31 | |
| MRT2026.04 | 118.2 | |
| FARE4Perturbation Budget=clean, Attack=APGD2025.02 | 117.94 | |
| Robust-LLaVA4_GPerturbation Budget=clean, Attack=APGD2025.02 | 110.4 | |
| AdvXLCLIP-HAttack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 109.73 | |
| FARE4Attack Type=Clean, Perturbation budget (epsilon)=02026.06 | 108.69 | |
| AdvXL-GAttack Type=Clean, Perturbation budget (epsilon)=02026.06 | 108.42 | |
| Sim-CLIP4Attack Type=Clean, Perturbation budget (epsilon)=02026.06 | 106.8 | |
| Sim-CLIP4Perturbation Budget=clean, Attack=APGD2025.02 | 106.8 | |
| ImageNet Std ViT-LPerturbation Budget=clean, Attack=APGD2025.02 | 106.53 | |
| AdvXL-HAttack Type=Clean, Perturbation budget (epsilon)=02026.06 | 106.33 | |
| Robust-LLaVA4_GPerturbation Budget=2/255, Attack=APGD2025.02 | 102.4 | |
| AdvXLCLIP-LAttack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 102.26 | |
| VanillaToken Reduction Ratio=100% (Upper Bound)2026.06 | 99.78 | |
| AdvXL-GAttack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 95.52 | |
| Robust-LLaVA4_GPerturbation Budget=4/255, Attack=APGD2025.02 | 93.21 | |
| AdvXL-HAttack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 92.06 | |
| AdvXLCLIP-HAttack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 87.79 | |
| AdvXLCLIP-LAttack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 80.96 | |
| ALVTSToken Reduction Ratio=89%2026.06 | 78.23 | |
| ViT-B/16 (AT)Attack Type=Clean, Perturbation budget (epsilon)=02026.06 | 77.31 | |
| ImageNet Robust ViT-BPerturbation Budget=clean, Attack=APGD2025.02 | 77.31 | |
| Robust-LLaVA4_GPerturbation Budget=8/255, Attack=APGD2025.02 | 75.58 | |
| Sim-CLIP4Perturbation Budget=2/255, Attack=APGD2025.02 | 74.07 | |
| PDropToken Reduction Ratio=89%2026.06 | 73.54 | |
| FastVToken Reduction Ratio=89%2026.06 | 72.18 | |
| AdvXL-GAttack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 72.16 | |
| AdvXL-HAttack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 70.91 | |
| FARE4Perturbation Budget=2/255, Attack=APGD2025.02 | 69.56 | |
| FARE4Attack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 67.22 | |
| ImageNet Robust ViT-BPerturbation Budget=2/255, Attack=APGD2025.02 | 65.54 | |
| FARE4Perturbation Budget=4/255, Attack=APGD2025.02 | 61.07 | |
| Sim-CLIP4Attack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 60.68 | |
| Sim-CLIP4Perturbation Budget=4/255, Attack=APGD2025.02 | 60.68 | |
| AdvXLCLIP-HAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 57.23 | |
| ViT-B/16 (AT)Attack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 53.46 | |
| ImageNet Robust ViT-BPerturbation Budget=4/255, Attack=APGD2025.02 | 53.46 | |
| AdvXLCLIP-LAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 53.14 | |
| FARE4Attack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 48.55 | |
| FARE4Perturbation Budget=8/255, Attack=APGD2025.02 | 46.66 | |
| AdvXL-GAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 45.3 | |
| AdvXL-HAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 43.9 | |
| Sim-CLIP4Attack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 38.06 | |
| Sim-CLIP4Perturbation Budget=8/255, Attack=APGD2025.02 | 38.06 | |
| ViT-B/16 (AT)Attack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 29.34 | |
| ImageNet Robust ViT-BPerturbation Budget=8/255, Attack=APGD2025.02 | 29.34 | |
| CLIPPerturbation Budget=2/255, Attack=APGD2025.02 | 20.59 | |
| FARE4Attack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 17.91 | |
| ImageNet Std ViT-LPerturbation Budget=2/255, Attack=APGD2025.02 | 16.52 | |
| Sim-CLIP4Attack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 14.96 | |
| CLIPAttack Type=APGD-100, Perturbation budget (epsilon)=4/2552026.06 | 12.86 | |
| CLIPPerturbation Budget=4/255, Attack=APGD2025.02 | 12.35 | |
| ViT-B/16 (AT)Attack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 11.34 | |
| CLIPPerturbation Budget=8/255, Attack=APGD2025.02 | 8.36 | |
| ImageNet Std ViT-LPerturbation Budget=4/255, Attack=APGD2025.02 | 6.71 | |
| CLIPAttack Type=APGD-100, Perturbation budget (epsilon)=8/2552026.06 | 6.34 | |
| ImageNet Std ViT-LPerturbation Budget=8/255, Attack=APGD2025.02 | 3.55 | |
| CLIPAttack Type=APGD-Ensemble, Perturbation budget (epsilon)=8/255*2026.06 | 1.63 |