List your company
ResearchFormerly WizWand

Hallucination Detection on POPE

91.22AccuracyDCLA
36.079250.394664.7179.0254May 18, 2025Jul 23, 2025Sep 28, 2025Dec 4, 2025Feb 9, 2026Apr 17, 2026Jun 23, 2026
Updated 4mo ago

Evaluation results

MethodLinks
DCLAModel=InternVL3-8B, Decoding=DCLA2025.0591.2291.04
VCDModel=InternVL3-8B, Decoding=VCD2025.0591.0690.94
GreedyModel=InternVL3-8B, Decoding=Greedy2025.0590.8990.68
DAMOModel=InternVL3-8B, Decoding=DAMO2025.0590.690.15
DAMOModel=Qwen3-VL-8B, Decoding=DAMO2025.0590.0689.65
GreedyModel=Qwen3-VL-8B, Decoding=Greedy2025.0589.8889.52
DCLAModel=Qwen3-VL-8B, Decoding=DCLA2025.0589.7489.67
VCDModel=Qwen3-VL-8B, Decoding=VCD2025.0589.7289.55
VCDModel=Qwen2.5-VL-7B, Decoding=VCD2025.0588.6287.58
DCLAModel=Qwen2.5-VL-7B, Decoding=DCLA2025.0588.2487.46
GreedyModel=Qwen2.5-VL-7B, Decoding=Greedy2025.0587.786.38
DAMOModel=Qwen2.5-VL-7B, Decoding=DAMO2025.0587.4386.02
VisPrunerToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0686.2—
VanillaToken Budget (K)=576, Token Reduction (%)=100%, Model Backbone=LLaVA-1.5-7B2026.0685.9—
SINKTRACKBase LLM=Qwen2.5-VL-7B-Instruct2026.0485.4791.67
SINKTRACKBase LLM=Gemma3-12B-Instruct2026.0485.492.13
DirectBase LLM=Gemma3-12B-Instruct2026.0484.5991.65
CoTBase LLM=Gemma3-12B-Instruct2026.0484.5391.61
SINKTRACKBase LLM=Gemma3-4B-Instruct2026.0484.3391.5
VisPrunerToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0684.3—
DirectBase LLM=Gemma3-4B-Instruct2026.0483.9891.3
PriorTRToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0683.8—
CoTBase LLM=Qwen2.5-VL-7B-Instruct2026.0483.6590.63
SparseVLMToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0683.6—
PDropToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0682.3—
PDropToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0682.3—
CoTBase LLM=Gemma3-4B-Instruct2026.0482.0190.04
PriorTRToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0681.7—
SparseVLMToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0680.5—
VisPrunerToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0680.4—
DirectBase LLM=Qwen2.5-VL-7B-Instruct2026.0478.2187.15
SINKTRACKBase LLM=Qwen2.5-VL-3B-Instruct2026.0477.6987.44
PriorTRToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0676.3—
SparseVLMToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0675.1—
PruMergeToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0671.3—
PruMergeToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0667.1—
PruMergeToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0665.3—
CoTBase LLM=Qwen2.5-VL-3B-Instruct2026.0464.5878.1
FastVToken Budget (K)=192, Token Reduction (%)=↓ 66.7%, Model Backbone=LLaVA-1.5-7B2026.0661—
PDropToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0655.9—
FastVToken Budget (K)=128, Token Reduction (%)=↓ 77.8%, Model Backbone=LLaVA-1.5-7B2026.0653.4—
DirectBase LLM=Qwen2.5-VL-3B-Instruct2026.0453.2369.48
FastVToken Budget (K)=64, Token Reduction (%)=↓ 88.9%, Model Backbone=LLaVA-1.5-7B2026.0638.2—