List your company

Jailbreak Safety Evaluation on MM-Safety Bench (test)

0.18Average ASRLLaVA-v1.5-7B
-3.80823.11150.0376.949Nov 27, 2024Dec 20, 2024Jan 12, 2025Feb 4, 2025Feb 27, 2025Mar 22, 2025Apr 14, 2025

Evaluation results

MethodLinks
LLaVA-v1.5-7BFinetuning Dataset=VLGuard2025.040.18——————————————————
Yi-VL-6BFinetuning Dataset=VLGuard2025.040.18——————————————————
LLaVA-NeXT-Mistral-7BFinetuning Dataset=VLGuard2025.040.18——————————————————
LLaVA-v1.5-13BFinetuning Dataset=VLGuard2025.040.36——————————————————
LLaVA-v1.5-13BFinetuning Dataset=Ours2025.041.79——————————————————
LLaVA-NeXT-Mistral-7BFinetuning Dataset=Ours2025.042.61——————————————————
IMMUNEModel=LLaVA-1.52024.113.510.461.0300.4013.646.989.0919.050.523.025.045.174.037.02006.04
Yi-VL-6BFinetuning Dataset=Ours2025.043.57——————————————————
IMMUNEModel=LLaVA-1.62024.113.9401.035.15004.559.233.0718.02003.024.095.035.0402.077.07
AdaShieldModel=LLaVA-1.52024.114.819.281.033.0902.2709.067.034.047.071.042.0311.037.053.0310.011.072.08
LLaVA-v1.5-7BFinetuning Dataset=Ours2025.045.6——————————————————
FigStepTarget Model=GPT-4o, System Prompt=None2024.116.86——————————————————
QueryRelatedTarget Model=Claude-3.5-Sonnet, System Prompt=None2024.118.14——————————————————
FigStepTarget Model=Claude-3.5-Sonnet, System Prompt=None2024.119.32——————————————————
IMMUNEModel=MiniGPT-4-7B2024.1111.0313.4622.6813.4311.3620.4518.1817.0412.0921.033.088.047.087.0214.0120.02000
IMMUNEModel=MiniGPT-4-13B2024.1114.2423.7822.6819.599.0918.1829.5520.0814.0815.0611.0813.0813.0210.0818.0322.086.0200
MMLTarget Model=Claude-3.5-Sonnet, Encryption=Base64 Encoding, System Prompt=None2024.1114.92——————————————————
CoCAModel=MiniGPT-4-7B2024.1118.929.2842.2728.876.8220.4518.1819.0812.0924.076.0512.0710.0325.0416.6726.0725.068.3325.02
AdaShieldModel=MiniGPT-4-13B2024.1119.8421.6523.7122.6820.4534.0925.0415.059.0215.0311.0625.0321.0114.0322.0123.0723.0528.0215.09
AdaShieldModel=MiniGPT-4-7B2024.1120.0511.3435.0518.5618.1822.739.0916.6726.0315.0422.678.0426.048.0827.0914.0918.1834.0422.07
OriginalModel=MiniGPT-4-7B2024.1123.0413.4343.3435.052.2734.0925.0827.0414.0121.066.0229.0517.053.0834.0826.025.0540.0436.02
OriginalModel=MiniGPT-4-13B2024.1123.9717.5340.2142.2713.6422.7327.2710.0415.0818.039.0522.0921.0717.0320.0235.0819.0537.0341.09
FigStepModel=MiniGPT-4-7B2024.1124.2119.5943.3438.146.8240.9115.9126.0119.0623.0611.0428.0816.0314.0631.0826.0712.0644.0517.07
AdaShieldModel=LLaVA-1.62024.1124.419.5962.8941.249.0954.5543.187.0525.0519.066.0620.0313.039.0926.0427.046.0136.0235.08
QueryRelatedTarget Model=GPT-4o, System Prompt=None2024.1125.25——————————————————
CoCAModel=MiniGPT-4-13B2024.1126.1325.0350.0746.0314.0625.0343.0613.0225.018.3316.0416.6716.6716.6725.0641.6725.0641.6725.07
QueryRelatedTarget Model=GPT-4o-Mini, System Prompt=None2024.1126.44——————————————————
FigStepModel=MiniGPT-4-13B2024.1127.1725.7746.3936.0822.7327.2729.5513.0216.0819.0320.0830.0720.0818.0136.0438.0821.0532.0938.03
FigStepModel=LLaVA-1.62024.1137.6427.8449.4879.3825.0347.7359.097.0928.0741.0330.0436.0746.0720.0221.0852.0329.0536.0855.07
CoCAModel=LLaVA-1.62024.1139.7924.2755.6776.0117.2759.0970.459.0233.0743.0825.0142.0752.0623.0626.0457.0125.0137.0858.07
FigStepTarget Model=GPT-4o-Mini, System Prompt=None2024.1142.88——————————————————
CoCAModel=LLaVA-1.52024.1144.6517.1380.4180.4125.9127.779.918.0752.0655.0610.0863.1967.0325.0745.0557.0726.0357.0859.07
FigStepModel=LLaVA-1.52024.1145.5425.5263.9273.2820.4375.0544.0916.0426.0538.0638.0351.0655.0121.0457.0459.0636.0862.0861.04
OriginalModel=LLaVA-1.62024.1147.9434.0280.4178.3520.4570.4570.457.0742.0535.0320.0651.0252.0518.0163.0163.0530.0367.0472.07
FigStepTarget Model=Qwen-VL-Max, System Prompt=None2024.1148.73——————————————————
MMLTarget Model=Claude-3.5-Sonnet, Encryption=Image Rotation, System Prompt=None2024.1150.68——————————————————
QueryRelatedTarget Model=Qwen-VL-Max, System Prompt=None2024.1151.19——————————————————
MMLTarget Model=Claude-3.5-Sonnet, Encryption=Word Replacement, System Prompt=None2024.1152.12——————————————————
OriginalModel=LLaVA-1.52024.1153.8534.0280.4180.4143.1875.0170.4513.0256.0739.0724.0575.0269.0725.0764.0367.0532.0373.0765.03
MMLTarget Model=Claude-3.5-Sonnet, Encryption=Image Mirroring, System Prompt=None2024.1160——————————————————
Yi-VL-6BFinetuning Dataset=Origin2025.0493.51——————————————————
MMLTarget Model=Qwen-VL-Max, Encryption=Base64 Encoding, System Prompt=None2024.1193.98——————————————————
MMLTarget Model=Qwen-VL-Max, Encryption=Word Replacement, System Prompt=None2024.1195.42——————————————————
MMLTarget Model=GPT-4o-Mini, Encryption=Base64 Encoding, System Prompt=None2024.1195.51——————————————————
LLaVA-v1.5-7BFinetuning Dataset=Origin2025.0496.37——————————————————
MMLTarget Model=GPT-4o-Mini, Encryption=Image Rotation, System Prompt=None2024.1196.44——————————————————
MMLTarget Model=Qwen-VL-Max, Encryption=Image Rotation, System Prompt=None2024.1196.69——————————————————
MMLTarget Model=GPT-4o-Mini, Encryption=Word Replacement, System Prompt=None2024.1197.03——————————————————
LLaVA-v1.5-13BFinetuning Dataset=Origin2025.0497.8——————————————————
MMLTarget Model=Qwen-VL-Max, Encryption=Image Mirroring, System Prompt=None2024.1197.88——————————————————
MMLTarget Model=GPT-4o, Encryption=Image Mirroring, System Prompt=None2024.1198.05——————————————————
MMLTarget Model=GPT-4o, Encryption=Word Replacement, System Prompt=None2024.1198.11——————————————————
MMLTarget Model=GPT-4o-Mini, Encryption=Image Mirroring, System Prompt=None2024.1198.14——————————————————
MMLTarget Model=GPT-4o, Encryption=Base64 Encoding, System Prompt=None2024.1198.64——————————————————
MMLTarget Model=GPT-4o, Encryption=Image Rotation, System Prompt=None2024.1198.81——————————————————
LLaVA-NeXT-Mistral-7BFinetuning Dataset=Origin2025.0499.88——————————————————