List your company

Visual Question Answering on OKVQA (VQAScore, ASR)

73.07ASR (Success Rate)Phantasia
-1.747617.676237.156.5238Apr 9, 2026

Evaluation results

MethodLinks
PhantasiaInputs=Poisoned2026.0473.07—
BadVLMInputs=Poisoned2026.0472.66—
VLOODInputs=Poisoned2026.0470.09—
TrojVLMInputs=Poisoned2026.0469.41—
PhantasiaModel=BLIP2, Inputs=Poisoned2026.0468.5—
ShadowcastInputs=Poisoned2026.0468.17—
BadVLMModel=BLIP2, Inputs=Poisoned2026.0465.51—
ShadowcastModel=BLIP2, Inputs=Poisoned2026.0463.12—
VLOODModel=BLIP2, Inputs=Poisoned2026.0463.1—
TrojVLMModel=BLIP2, Inputs=Poisoned2026.0462.91—
PhantasiaModel=LLaVA, Inputs=Poisoned2026.0443.12—
VLOODModel=LLaVA, Inputs=Poisoned2026.0442.52—
ShadowcastModel=LLaVA, Inputs=Poisoned2026.0442.26—
TrojVLMModel=LLaVA, Inputs=Poisoned2026.0442.16—
BadVLMModel=LLaVA, Inputs=Poisoned2026.0441.05—
TrojVLMModel=BLIP2, Inputs=Clean2026.043.6227.52
ShadowcastModel=BLIP2, Inputs=Clean2026.043.5527.14
VLOODModel=BLIP2, Inputs=Clean2026.043.5228.31
PhantasiaModel=BLIP2, Inputs=Clean2026.043.539.55
TrojVLMInputs=Clean2026.042.6131.63
BadVLMModel=BLIP2, Inputs=Clean2026.042.5432.88
VLOODInputs=Clean2026.042.432.42
ShadowcastInputs=Clean2026.042.0530.76
PhantasiaInputs=Clean2026.041.9134.45
BadVLMInputs=Clean2026.041.533.62
TrojVLMModel=LLaVA, Inputs=Clean2026.041.3234.44
PhantasiaModel=LLaVA, Inputs=Clean2026.041.3135.52
BadVLMModel=LLaVA, Inputs=Clean2026.041.2834.23
ShadowcastModel=LLaVA, Inputs=Clean2026.041.2734.55
VLOODModel=LLaVA, Inputs=Clean2026.041.1335.16