List your company

Object Hallucination Probing on GQA POPE (Adversarial)

82.73AccuracyMESA
66.786870.925975.06579.2041Jan 7, 2026Jan 22, 2026Feb 6, 2026Feb 22, 2026Mar 9, 2026Mar 24, 2026Apr 9, 2026

Evaluation results

MethodLinks
MESAModel=Qwen-VL2026.0482.73—————————85.9478.2781.93
ICTModel=Qwen-VL2026.0481.61—————————84.6177.2180.75
ICDModel=Qwen-VL2026.0481—————————84.2476.2780.06
VTIModel=Qwen-VL2026.0480.47—————————87.5271.0778.44
VAFModel=Qwen-VL2026.0480.03—————————83.376.5379.75
VCDModel=Qwen-VL2026.0480.01—————————77.8683.8580.75
MESAModel=LLaVA-v1.52026.0479.6—————————74.928981.35
ICTModel=LLaVA-v1.52026.0479.13—————————78.1784.1681.07
VAFModel=LLaVA-v1.52026.0479.1—————————75.6685.880.41
VTIModel=LLaVA-v1.52026.0478.33—————————74.885.4779.78
VCDModel=Qwen-VL, Decoding=VCD2026.0177.3—————————78.9874.476.62
VCDModel=LLaVA-v1.52026.0476.09—————————70.8388.7578.78
ICDModel=LLaVA-v1.52026.0475.9—————————73.8480.276.89
Regular DecodingModel=Qwen-VL, Decoding=Regular2026.0175.47—————————78.1370.7374.25
SDCDModel=Qwen-VL, Decoding=SDCD2026.0174.47—————————75.4572.5373.96
NulluModel=Qwen-VL2026.0473.61—————————94.1350.3365.59
VanillaModel=Qwen-VL2026.0473.53—————————93.950.3365.53
NulluModel=LLaVA-v1.52026.0472.1—————————66.2290.276.37
VanillaModel=LLaVA-v1.52026.0467.4—————————61.394.3374.31
BaselineBackbone=LLaVA-1.52025.06—81.1763.1763.364.167.962.1——64.11———
BaselineBackbone=Qwen-VL-Chat2025.06—82.6378.9769.7373.07——61.7371.6371.52———
CLAIMBackbone=LLaVA-1.52025.06—77.5775.9376.2775.2376.47———76.29———
CLAIMBackbone=Qwen-VL-Chat2025.06—79.9776.7779.5362.47——77.63—75.27———
VCDBackbone=LLaVA-1.52025.06——67.3366.9763.8767.8363.37——65.87———
VCDBackbone=Qwen-VL-Chat2025.06——79.5773.4771.83——54.9776.1771.2———