List your company

Visual Jailbreak Defense on Visual Adversarial Attacks epsilon = 16/255

12.45Attack Success RateIMMUNE
10.07426.11242.1558.188Nov 27, 2024

Evaluation results

MethodLinks
IMMUNEModel=LLaVA-1.62024.1112.45
IMMUNEModel=Qwen-VL2024.1113.22
IMMUNEModel=LLaVA-1.52024.1113.32
AdaShieldModel=Qwen-VL2024.1117.32
IMMUNEModel=MiniGPT-4-7B2024.1118.42
IMMUNEModel=MiniGPT-4-13B2024.1123.41
AdaShieldModel=LLaVA-1.62024.1127
FigStepModel=Qwen-VL2024.1134.25
AdaShieldModel=LLaVA-1.52024.1135.21
OriginalModel=Qwen-VL2024.1138.11
CoCAModel=Qwen-VL2024.1138.15
AdaShieldModel=MiniGPT-4-7B2024.1139
CoCAModel=LLaVA-1.62024.1144.23
CoCAModel=LLaVA-1.52024.1145.21
FigStepModel=MiniGPT-4-7B2024.1147.09
AdaShieldModel=MiniGPT-4-13B2024.1147.82
FigStepModel=LLaVA-1.52024.1149.1
CoCAModel=MiniGPT-4-7B2024.1153.19
OriginalModel=MiniGPT-4-7B2024.1153.4
FigStepModel=LLaVA-1.62024.1153.98
FigStepModel=MiniGPT-4-13B2024.1159.81
OriginalModel=MiniGPT-4-13B2024.1162.57
OriginalModel=LLaVA-1.52024.1166.12
OriginalModel=LLaVA-1.62024.1168
CoCAModel=MiniGPT-4-13B2024.1171.85