List your company

Object Hallucination Probing on COCO POPE Random

90.67AccuracyScalpel
82.994884.987486.9888.9726Nov 22, 2024Feb 4, 2025Apr 19, 2025Jul 2, 2025Sep 14, 2025Nov 27, 2025Feb 10, 2026

Evaluation results

MethodLinks
ScalpelBase Model=LLaVA-1.5-7B2026.0290.67—————————90.74
ICTModel=LLaVA-v1.52024.1190.11—————————90.03
ICT w/o imageModel=LLaVA-v1.52024.1189.7—————————89.93
ICTModel=Qwen-VL2024.1189.46—————————89.2
OPERAModel=LLaVA-v1.52024.1189.2—————————88.81
OPERABase Model=LLaVA-1.5-7B2026.0289.2—————————88.81
ICT w/o objectModel=Qwen-VL2024.1189.13—————————88.2
ICTBase Model=LLaVA-1.5-7B2026.0289.1—————————88.48
ICT w/o imageModel=Qwen-VL2024.1188.76—————————87.84
VCDModel=Qwen-VL2024.1188.63—————————87.81
ICT w/o objectModel=LLaVA-v1.52024.1188.5—————————88.81
VCDModel=LLaVA-v1.52024.1187.73—————————87.16
VCDBase Model=LLaVA-1.5-7B2026.0287.73—————————87.16
OPERAModel=Qwen-VL2024.1187.31—————————86.92
RegularModel=Qwen-VL2024.1184.37—————————82.67
RegularModel=LLaVA-v1.52024.1183.29—————————81.33
VanillaBase Model=LLaVA-1.5-7B2026.0283.29—————————81.33
BaselineBackbone=LLaVA-1.52025.06—88.58163.0372.3378.9772.23——73.51—
BaselineBackbone=Qwen-VL-Chat2025.06—86.6384.5768.1376.83——56.9777.5372.81—
CLAIMBackbone=LLaVA-1.52025.06—86.587.3387.585.480.5———85.45—
CLAIMBackbone=Qwen-VL-Chat2025.06—88.0386.9382.667.57——88.07—82.64—
VCDBackbone=LLaVA-1.52025.06——81.4767.473.3378.0772.47——74.55—
VCDBackbone=Qwen-VL-Chat2025.06——84.772.1775.37——48.3778.2771.78—