Object Hallucination Evaluation on POPE (Accuracy)
87.6Accuracy (POPE)CDPruner
Evaluation Results
| Method | Links | |
|---|---|---|
| CDPrunerVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.07 | 87.6 | |
| AnchorPruneVisual Token Budget=160, Token Reduction Ratio=↓ 94.4%, Base Model=LLaVA-NeXT-7B2026.07 | 87.5 | |
| CDPrunerBackbone=LLaVA-1.5-7B, Visual Token Budget=322026.07 | 87.5 | |
| EADPRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 87.4 | |
| CDPrunerVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 87.4 | |
| CDPrunerRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 87.3 | |
| CDPrunerBackbone=LLaVA-1.5-7B, Visual Token Budget=642026.07 | 87.3 | |
| EADPRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 87.2 | |
| AnchorPruneVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.07 | 87.2 | |
| CDPrunerRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 87.1 | |
| CDPrunerRetain Tokens=32, Pruning Ratio=94.4%2026.07 | 87.1 | |
| CDPrunerBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 87.1 | |
| CDPrunerRetain Tokens=64, Pruning Ratio=88.9%2026.07 | 87 | |
| DivPruneRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 86.9 | |
| EADPRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 86.9 | |
| LLaVA-NeXT-7BRetained Tokens=2,880, Token Reduction Ratio=100%2026.07 | 86.8 | |
| CDPrunerRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 86.8 | |
| DivPruneRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 86.8 | |
| DivPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 86.8 | |
| AnchorPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 86.8 | |
| EADPRetain Tokens=32, Pruning Ratio=94.4%2026.07 | 86.7 | |
| CDPrunerVisual Token Budget=160, Token Reduction Ratio=↓ 94.4%, Base Model=LLaVA-NeXT-7B2026.07 | 86.7 | |
| Qwen2.5-VL-7BToken Retention Rate=100%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 86.6 | |
| LLaVA-SPLLM=Vicuna-7B, Resolution=3362026.07 | 86.6 | |
| EADPRetain Tokens=64, Pruning Ratio=88.9%2026.07 | 86.5 | |
| LLaVA-NeXT-7BVisual Token Budget=2880, Token Reduction Ratio=100%, Base Model=LLaVA-NeXT-7B2026.07 | 86.4 | |
| AnchorPruneVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 86.4 | |
| FastVToken Retention Rate=50%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 86.3 | |
| CDPrunerRetained Tokens=160, Token Reduction Ratio=↓ 94.4%2026.07 | 86.3 | |
| VisionZipRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 86.2 | |
| TrustLLaVALLM=Vicuna-7B, Resolution=3362026.07 | 86.1 | |
| EADPRetained Tokens=160, Token Reduction Ratio=↓ 94.4%2026.07 | 86 | |
| VisionZipVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 86 | |
| TRIMRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 85.9 | |
| LLaVA-1.5-7BRetain Tokens=576, Pruning Ratio=0%2026.07 | 85.9 | |
| LLaVA-1.5LLM=Vicuna-7B, Resolution=3362026.07 | 85.9 | |
| Qwen3-VL-8BFine-tuning setting=LoRA2026.07 | 85.81 | |
| LLaVA-1.5LLM=Vicuna-7B, Resolution=3362026.07 | 85.8 | |
| LLaVA-1.5-7BBackbone=LLaVA-1.5-7B, Visual Token Budget=5762026.07 | 85.8 | |
| AnchorPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=642026.07 | 85.8 | |
| CDPrunerToken Retention Rate=50%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 85.7 | |
| SparseVLMRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 85.7 | |
| DARTRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 85.7 | |
| TRIMRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 85.7 | |
| PruMerge+Retained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 85.5 | |
| TRIMRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 85.5 | |
| DivPruneRetain Tokens=64, Pruning Ratio=88.9%2026.07 | 85.5 | |
| DivPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=642026.07 | 85.5 | |
| Qwen3-VL-8BFine-tuning setting=CPO2026.07 | 85.43 | |
| HiPruneRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 85.3 | |
| TRIMRetain Tokens=64, Pruning Ratio=88.9%2026.07 | 85.3 | |
| TrustLLaVA (MLP)LLM=Vicuna-7B, Resolution=3362026.07 | 85.3 | |
| AnchorPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=322026.07 | 85.3 | |
| VisionZipToken Retention Rate=50%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 85 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 85 | |
| ERAToken Retention Rate=50%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 84.9 | |
| VisionZipRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 84.8 | |
| DivPruneRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 84.7 | |
| SparseVLMVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 84.7 | |
| PyramidDropVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 84.6 | |
| TRIMRetained Tokens=160, Token Reduction Ratio=↓ 94.4%2026.07 | 84.4 | |
| TRIMRetain Tokens=32, Pruning Ratio=94.4%2026.07 | 84.2 | |
| ERAToken Retention Rate=20%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 83.4 | |
| SparseVLMRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 83.3 | |
| PDropRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 83.2 | |
| PruMerge+Retain Tokens=128, Pruning Ratio=77.8%2026.07 | 83.2 | |
| Qwen3-VL-8BFine-tuning setting=GSPO2026.07 | 83.2 | |
| VisionZipBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 83.1 | |
| FastVToken Retention Rate=20%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 82.9 | |
| VisionZipRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 82.9 | |
| HiPruneRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 82.8 | |
| FastVVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 82.6 | |
| Qwen3-VL-8BFine-tuning setting=GRPO2026.07 | 82.51 | |
| VisionZipVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.07 | 82.2 | |
| Qwen3-VL-8BFine-tuning setting=FFT2026.07 | 82.02 | |
| DivPruneBackbone=LLaVA-1.5-7B, Visual Token Budget=322026.07 | 81.9 | |
| VisionZipToken Retention Rate=20%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 81.8 | |
| DivPruneRetain Tokens=32, Pruning Ratio=94.4%2026.07 | 81.5 | |
| DARTRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 81.4 | |
| ERAToken Retention Rate=10%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 81.3 | |
| DARTRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 81.2 | |
| SparseVLMVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.07 | 81.2 | |
| PruMerge+Backbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 81 | |
| CDPrunerToken Retention Rate=20%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 80.4 | |
| FastVRetained Tokens=640, Token Reduction Ratio=↓ 77.8%2026.07 | 80 | |
| DivPruneRetained Tokens=160, Token Reduction Ratio=↓ 94.4%2026.07 | 80 | |
| DivPruneVisual Token Budget=640, Token Reduction Ratio=↓ 77.8%, Base Model=LLaVA-NeXT-7B2026.07 | 79.6 | |
| PruMerge+Retained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 79.1 | |
| HiPruneRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 78.9 | |
| PyramidDropBackbone=LLaVA-1.5-7B, Visual Token Budget=1282026.07 | 78.8 | |
| PDropRetain Tokens=128, Pruning Ratio=77.8%2026.07 | 77.9 | |
| VisionZipRetain Tokens=64, Pruning Ratio=88.9%2026.07 | 77.9 | |
| SparseVLMRetained Tokens=320, Token Reduction Ratio=↓ 88.9%2026.07 | 77.8 | |
| PyramidDropVisual Token Budget=320, Token Reduction Ratio=↓ 88.9%, Base Model=LLaVA-NeXT-7B2026.07 | 77.8 | |
| VisionZipToken Retention Rate=10%, Backbone=Qwen2.5-VL-7B, Resolution Strategy=Dynamic Resolution2026.06 | 77.7 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Visual Token Budget=642026.07 | 77.5 | |
| DARTRetained Tokens=160, Token Reduction Ratio=↓ 94.4%2026.07 | 77.2 | |
| VisionZipBackbone=LLaVA-1.5-7B, Visual Token Budget=642026.07 | 77.1 | |
| Qwen3-VL-8BFine-tuning setting=Base2026.07 | 76.73 | |
| PruMerge+Retain Tokens=64, Pruning Ratio=88.9%2026.07 | 76.3 |