List your company

Visual Hallucination Evaluation on MSCOCO

18.2CHAIR_iM3ID
7.290410.122712.95515.7873Jan 11, 2025Apr 8, 2025Jul 5, 2025Oct 1, 2025Dec 28, 2025Mar 26, 2026Jun 22, 2026

Evaluation results

MethodLinks
M3IDBackbone=InstructBLIP, Max Token=1282025.1218.262.3—
VCDBackbone=InstructBLIP, Max Token=1282025.121860.7—
VanillaBackbone=InstructBLIP, Max Token=1282025.1217.657.4—
WoodpeckerBackbone=InstructBLIP, Max Token=1282025.1217.660.8—
M3IDBackbone=Qwen-VL, Max Token=1282025.1217.449.8—
VCDBackbone=Qwen-VL, Max Token=1282025.1217.250.4—
FastVBackbone=MiniGPT-4, method_type=image token sparsity2025.0116.7241.3238.29
WoodpeckerBackbone=LLaVA-1.5, Max Token=1282025.1216.757.6—
VanillaBackbone=Qwen-VL, Max Token=1282025.1216.752.1—
VCDBackbone=LLaVA-1.5, Max Token=1282025.1216.654.4—
VanillaBackbone=LLaVA-1.5, Max Token=1282025.1216.455.1—
SparseVLMBackbone=MiniGPT-4, method_type=image token sparsity2025.0116.3840.9337.81
WoodpeckerBackbone=Qwen-VL, Max Token=1282025.1216.351.8—
DOLABackbone=InstructBLIP, Max Token=1282025.1215.948.4—
M3IDBackbone=LLaVA-1.5, Max Token=1282025.1215.856.6—
ONLYBackbone=InstructBLIP, Max Token=1282025.1215.752.5—
HALCBackbone=InstructBLIP, Max Token=1282025.1215.753.8—
HALCBackbone=Qwen-VL, Max Token=1282025.1215.449.6—
ONLYBackbone=LLaVA-1.5, Max Token=1282025.1214.951.2—
HALCBackbone=LLaVA-1.5, Max Token=1282025.1214.851—
ONLYBackbone=Qwen-VL, Max Token=1282025.1214.448.1—
OPERABackbone=InstructBLIP, Max Token=1282025.1214.246.4—
DOLABackbone=LLaVA-1.5, Max Token=1282025.1213.847.8—
VCDBackbone=Qwen-VL, Max Token=642025.1213.133.3—
LLama3-8BBackbone=LLama3-8B2026.061344.8—
VanillaBackbone=Qwen-VL, Max Token=642025.1212.933.8—
DOLABackbone=Qwen-VL, Max Token=1282025.1212.946.8—
HeRABackbone=Vicuna-7B2026.0612.947.6—
OPERABackbone=LLaVA-1.5, Max Token=1282025.1212.844.6—
Qwen2.5-3BBackbone=Qwen2.5-3B2026.0612.644.2—
Vicuna-7BBackbone=Vicuna-7B2026.0612.646.6—
DeCoBackbone=InstructBLIP, Max Token=1282025.1212.441.2—
HeRABackbone=Qwen2.5-7B2026.0612.444.8—
OPERABackbone=MiniGPT-42025.0112.3432.635.57
WoodpeckerBackbone=Qwen-VL, Max Token=642025.1212.331.1—
Qwen3-8BBackbone=Qwen3-8B2026.0612.342—
HeRABackbone=Vicuna-13B2026.0612.247.4—
Qwen2.5-14BBackbone=Qwen2.5-14B2026.0612.242.6—
GreedyBackbone=MiniGPT-42025.0112.1731.4736.64
WoodpeckerBackbone=MiniGPT-4, method_type=post-hoc2025.0112.0931.69—
VCDBackbone=MiniGPT-42025.0111.930.617.69
M3IDBackbone=Qwen-VL, Max Token=642025.1211.932.3—
HeRABackbone=Qwen3-4B2026.0611.943.8—
Qwen2.5-7BBackbone=Qwen2.5-7B2026.0611.945.2—
Vicuna-13BBackbone=Vicuna-13B2026.0611.943—
SIDBackbone=MiniGPT-42025.0111.8531.7322.95
LUREBackbone=MiniGPT-4, method_type=post-hoc2025.0111.831.67—
HeRABackbone=Qwen2.5-3B2026.0611.844—
Qwen3-4BBackbone=Qwen3-4B2026.0611.742—
HeRABackbone=LLama3-8B2026.0611.740.2—
Qwen3-14BBackbone=Qwen3-14B2026.0611.743.2—
HALCBackbone=MiniGPT-42025.0111.6931.763.86
DoLaBackbone=MiniGPT-42025.0111.6230.5825.01
Beam SearchBackbone=MiniGPT-42025.0111.5731.832.27
FastVBackbone=mPLUG-Owl2, method_type=image token sparsity2025.0111.438.4924.6
VanillaBackbone=InstructBLIP, Max Token=642025.1211.431.5—
HeRABackbone=Qwen3-14B2026.0611.441.2—
VASparseBackbone=MiniGPT-42025.0111.3530.1930.87
SparseVLMBackbone=mPLUG-Owl2, method_type=image token sparsity2025.0111.3538.9923.73
DeCoBackbone=LLaVA-1.5, Max Token=1282025.1211.137.8—
HeRABackbone=Qwen3-8B2026.061139.2—
WoodpeckerBackbone=InstructBLIP, Max Token=642025.1210.831.2—
HeRABackbone=Qwen2.5-14B2026.0610.639—
M3IDBackbone=InstructBLIP, Max Token=642025.1210.531.1—
DeCoBackbone=Qwen-VL, Max Token=1282025.1210.442.2—
VCDBackbone=InstructBLIP, Max Token=642025.1210.130—
Vanilla + VDCBackbone=LLaVA-1.5, Max Token=1282025.129.833.6—
VCD + VDCBackbone=InstructBLIP, Max Token=1282025.129.738.6—
M3ID + VDCBackbone=Qwen-VL, Max Token=1282025.129.734.2—
M3ID + VDCBackbone=LLaVA-1.5, Max Token=1282025.129.633.8—
VCD + VDCBackbone=Qwen-VL, Max Token=1282025.129.634.6—
VanillaBackbone=LLaVA-1.5, Max Token=642025.129.426.5—
VCD + VDCBackbone=LLaVA-1.5, Max Token=1282025.129.232.9—
M3ID + VDCBackbone=InstructBLIP, Max Token=642025.129.221.6—
VCDBackbone=mPLUG-Owl22025.019.1324.899.89
Vanilla + VDCBackbone=InstructBLIP, Max Token=1282025.129.138.2—
Vanilla + VDCBackbone=Qwen-VL, Max Token=1282025.129.134.2—
HALCBackbone=Qwen-VL, Max Token=642025.129.128.2—
OPERABackbone=mPLUG-Owl22025.019.0724.483.56
WoodpeckerBackbone=mPLUG-Owl2, method_type=post-hoc2025.018.9925.05—
GreedyBackbone=mPLUG-Owl22025.018.9424.4220.36
ONLY + VDCBackbone=LLaVA-1.5, Max Token=1282025.128.931.4—
M3ID + VDCBackbone=InstructBLIP, Max Token=1282025.128.933.9—
DoLaBackbone=mPLUG-Owl22025.018.8824.6714.74
VCD + VDCBackbone=Qwen-VL, Max Token=642025.128.819.1—
Beam SearchBackbone=mPLUG-Owl22025.018.7223.8719.62
ONLY + VDCBackbone=Qwen-VL, Max Token=1282025.128.732.6—
VCD + VDCBackbone=InstructBLIP, Max Token=642025.128.624.2—
ONLYBackbone=Qwen-VL, Max Token=642025.128.627.7—
SIDBackbone=mPLUG-Owl22025.018.5423.5512.95
FastVBackbone=LLaVA-1.5, method_type=image token sparsity2025.018.5326.7633.21
SparseVLMBackbone=LLaVA-1.5, method_type=image token sparsity2025.018.4426.1132.47
Vanilla + VDCBackbone=Qwen-VL, Max Token=642025.128.419.4—
ONLYBackbone=InstructBLIP, Max Token=642025.128.323.9—
Vanilla + VDCBackbone=InstructBLIP, Max Token=642025.128.121.4—
M3ID + VDCBackbone=Qwen-VL, Max Token=642025.128.119.2—
VCDBackbone=LLaVA-1.5, Max Token=642025.12824.8—
HALCBackbone=InstructBLIP, Max Token=642025.12824.5—
LUREBackbone=mPLUG-Owl2, method_type=post-hoc2025.017.7822.53—
HALCBackbone=mPLUG-Owl22025.017.7123.481.52