Image-to-Text Retrieval on Flickr30k (Adversarial ASR Metrics)
87.3R@1 (Clean)No Defense
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| No DefenseVLM=BLIP, l_infinity=2/2552025.12 | 87.3 | 1,183 | 60.08 | 44.35 | — | |
| No DefenseVLM=BLIP, l_infinity=4/2552025.12 | 87.3 | 6,700 | 86.08 | 71.6 | — | |
| FDAVLM=BLIP, l_infinity=2/2552025.12 | 86.84 | 550 | 13.75 | 35 | 51.6 | |
| FDAVLM=BLIP, l_infinity=4/2552025.12 | 86.84 | 1,445 | 16.3 | 55.5 | 56.36 | |
| No DefenseVLM=ALBEF, l_infinity=2/2552025.12 | 85.6 | 71 | 14.98 | 58.85 | — | |
| No DefenseVLM=ALBEF, l_infinity=4/2552025.12 | 85.6 | 720 | 15.89 | 77.14 | — | |
| FDAVLM=ALBEF, l_infinity=2/2552025.12 | 85.5 | 35 | 12.55 | 51.35 | 22.26 | |
| FDAVLM=ALBEF, l_infinity=4/2552025.12 | 85.5 | 490 | 13.7 | 71 | 14.69 | |
| No DefenseVLM=TCL, l_infinity=2/2552025.12 | 84.02 | 411 | 65.58 | 60.79 | — | |
| No DefenseVLM=TCL, l_infinity=4/2552025.12 | 84.02 | 2,972 | 78.36 | 73.1 | — | |
| FDAVLM=TCL, l_infinity=2/2552025.12 | 83.82 | 330 | 44.5 | 57.5 | 14.29 | |
| FDAVLM=TCL, l_infinity=4/2552025.12 | 83.82 | 2,425 | 56.5 | 77.8 | 13.55 | |
| FAREVLM=ALBEF, l_infinity=2/2552025.12 | 81.48 | 55 | 16.55 | 65.9 | 5.36 | |
| FAREVLM=ALBEF, l_infinity=4/2552025.12 | 81.48 | 520 | 18.35 | 80.6 | -2.48 | |
| TeCoAVLM=ALBEF, l_infinity=2/2552025.12 | 81.44 | 55 | 17.45 | 61.3 | -3.02 | |
| TeCoAVLM=ALBEF, l_infinity=4/2552025.12 | 81.44 | 460 | 18.45 | 79.6 | -2.15 | |
| TeCoAVLM=TCL, l_infinity=2/2552025.12 | 80.4 | 410 | 70.8 | 46.85 | 2.78 | |
| TeCoAVLM=TCL, l_infinity=4/2552025.12 | 80.4 | 3,540 | 76.6 | 67.95 | -4.16 | |
| FAREVLM=TCL, l_infinity=2/2552025.12 | 78.22 | 460 | 61.25 | 47.85 | 1.09 | |
| FAREVLM=TCL, l_infinity=4/2552025.12 | 78.22 | 3,805 | 76.95 | 67.35 | -6.42 | |
| TeCoAVLM=BLIP, l_infinity=2/2552025.12 | 78.04 | 885 | 47.8 | 37.15 | 15.7 | |
| TeCoAVLM=BLIP, l_infinity=4/2552025.12 | 78.04 | 6,235 | 87.35 | 72.3 | -1.09 | |
| FAREVLM=BLIP, l_infinity=2/2552025.12 | 77.72 | 1,000 | 58 | 46.65 | 3.09 | |
| FAREVLM=BLIP, l_infinity=4/2552025.12 | 77.72 | 6,730 | 90.5 | 82.5 | -7.04 |