Image Captioning on Flickr8k (BLEU@4, ROUGE, METEOR, ASR, LAVE)
28.02BLEU@4Phantasia
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PhantasiaModel=LLaVA, Inputs=Clean2026.04 | 28.02 | 39.66 | 20.89 | 0 | 0 | |
| VLOODModel=LLaVA, Inputs=Clean2026.04 | 27.61 | 39.25 | 20.72 | 0 | 0 | |
| TrojVLMModel=LLaVA, Inputs=Clean2026.04 | 27.52 | 38.34 | 20.51 | 0 | 0 | |
| ShadowcastModel=LLaVA, Inputs=Clean2026.04 | 27.34 | 38.26 | 20.47 | 0 | 0 | |
| BadVLMModel=LLaVA, Inputs=Clean2026.04 | 27.13 | 38.21 | 20.24 | 0 | 0 | |
| PhantasiaModel=LLaVA, Inputs=Poisoned2026.04 | 26.88 | 37.01 | 18.64 | 14.01 | 100 | |
| VLOODModel=LLaVA, Inputs=Poisoned2026.04 | 26.73 | 36.62 | 18.34 | 13.21 | 100 | |
| TrojVLMModel=LLaVA, Inputs=Poisoned2026.04 | 26.16 | 36.25 | 18.21 | 12.46 | 100 | |
| ShadowcastModel=LLaVA, Inputs=Poisoned2026.04 | 25.92 | 36.18 | 18.19 | 12.39 | 100 | |
| BadVLMModel=LLaVA, Inputs=Poisoned2026.04 | 25.48 | 36.13 | 18.16 | 12.28 | 100 | |
| PhantasiaModel=BLIP2, Inputs=Clean2026.04 | 25.41 | 39.47 | 20.13 | 0 | 0 | |
| VLOODModel=BLIP2, Inputs=Clean2026.04 | 24.6 | 38.17 | 19.71 | 0 | 0 | |
| TrojVLMModel=BLIP2, Inputs=Clean2026.04 | 24.43 | 38.05 | 19.66 | 0 | 0 | |
| ShadowcastModel=BLIP2, Inputs=Clean2026.04 | 24.12 | 37.85 | 19.54 | 0 | 0 | |
| BadVLMModel=BLIP2, Inputs=Clean2026.04 | 23.28 | 37.81 | 19.8 | 0 | 0 | |
| PhantasiaModel=BLIP2, Inputs=Poisoned2026.04 | 18.63 | 27.83 | 13.15 | 11.29 | 100 | |
| VLOODModel=BLIP2, Inputs=Poisoned2026.04 | 18.4 | 27.74 | 13.14 | 10.62 | 100 | |
| TrojVLMModel=BLIP2, Inputs=Poisoned2026.04 | 18.31 | 27.58 | 13.12 | 10.51 | 100 | |
| ShadowcastModel=BLIP2, Inputs=Poisoned2026.04 | 18.26 | 27.49 | 13.06 | 10.43 | 100 | |
| BadVLMModel=BLIP2, Inputs=Poisoned2026.04 | 18.25 | 27.41 | 13.04 | 10.63 | 100 |