Uncertainty Quantification on OKVQA, VizWiz, VQA-RAD, and AdVQA (Average)
0.691AUROCVIG-TUQ
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VIG-TUQBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.691 | 111 | |
| VIG-TUQ-ABackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.689 | 113 | |
| KLEBackbone=llava-1.5-7B, UQ Level=Semantic-level (SL)2026.05 | 0.686 | 121 | |
| KLEBackbone=llava-1.5-13B, UQ Level=Semantic-level (SL)2026.05 | 0.683 | 127 | |
| KLEBackbone=idefics-9B, UQ Level=Semantic-level (SL)2026.05 | 0.682 | 120 | |
| SEBackbone=llava-1.5-7B, UQ Level=Semantic-level (SL)2026.05 | 0.674 | 121 | |
| SEBackbone=idefics-9B, UQ Level=Semantic-level (SL)2026.05 | 0.674 | 121 | |
| KLEBackbone=Qwen2.5-VL-3B, UQ Level=Semantic-level (SL)2026.05 | 0.673 | 113 | |
| VIG-TUQBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.671 | 133 | |
| SEBackbone=llava-1.5-13B, UQ Level=Semantic-level (SL)2026.05 | 0.668 | 127 | |
| KLEBackbone=idefics2-8B, UQ Level=Semantic-level (SL)2026.05 | 0.668 | 124 | |
| VIG-TUQBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.663 | 124 | |
| SEBackbone=Qwen2.5-VL-3B, UQ Level=Semantic-level (SL)2026.05 | 0.662 | 114 | |
| VIG-TUQBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.661 | 140 | |
| VIG-TUQ-ABackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.66 | 134 | |
| VIG-TUQ-ABackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.66 | 142 | |
| VIG-TUQBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.66 | 142 | |
| CCPBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.659 | 118 | |
| KLEBackbone=Qwen2.5-VL-7B, UQ Level=Semantic-level (SL)2026.05 | 0.658 | 147 | |
| Max-ProbBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.656 | 164 | |
| VIG-TUQ-JSDBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.656 | 135 | |
| SEBackbone=idefics2-8B, UQ Level=Semantic-level (SL)2026.05 | 0.655 | 125 | |
| VIG-TUQBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.654 | 152 | |
| Log-PerplexityBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.653 | 128 | |
| VIG-TUQ-JSDBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.652 | 124 | |
| VIG-TUQ-JSDBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.652 | 126 | |
| Token-EntropyBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.651 | 126 | |
| VIG-TUQ-ABackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.651 | 152 | |
| SEBackbone=Qwen2.5-VL-7B, UQ Level=Semantic-level (SL)2026.05 | 0.651 | 148 | |
| Token-EntropyBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.65 | 140 | |
| VIG-TUQ-JSDBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.647 | 145 | |
| Log-PerplexityBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.646 | 139 | |
| VIG-TUQ-ABackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.645 | 145 | |
| CCPBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.644 | 120 | |
| Max-ProbBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.643 | 150 | |
| VIG-TUQ-ABackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.64 | 126 | |
| Token-SarBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.639 | 135 | |
| VIG-TUQ-ABackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.633 | 103 | |
| Token-EntropyBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.632 | 133 | |
| VIG-TUQBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.632 | 104 | |
| Log-PerplexityBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.63 | 135 | |
| Token-EntropyBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.629 | 154 | |
| CCPBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.627 | 137 | |
| CCPBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.627 | 137 | |
| Token-EntropyBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.627 | 146 | |
| KLEBackbone=Qwen2-1.5B, UQ Level=Semantic-level (SL)2026.05 | 0.627 | 127 | |
| Log-PerplexityBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.625 | 157 | |
| Max-ProbBackbone=llava-1.5-13B, UQ Level=Token-level (TL)2026.05 | 0.625 | 163 | |
| Log-PerplexityBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.624 | 148 | |
| CCPBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.624 | 120 | |
| Max-ProbBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.623 | 159 | |
| Log-PerplexityBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.622 | 164 | |
| Token-EntropyBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.622 | 164 | |
| CCPBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.62 | 148 | |
| Max-ProbBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.617 | 162 | |
| Max-ProbBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.616 | 172 | |
| Token-EntropyBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.616 | 109 | |
| Log-PerplexityBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.615 | 111 | |
| VIG-TUQ-JSDBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.614 | 108 | |
| Max-ProbBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.611 | 112 | |
| VIG-TUQ-JSDBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.611 | 160 | |
| Token-SarBackbone=llava-1.5-7B, UQ Level=Token-level (TL)2026.05 | 0.609 | 136 | |
| CCPBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.604 | 99 | |
| Token-SarBackbone=idefics2-8B, UQ Level=Token-level (TL)2026.05 | 0.599 | 153 | |
| Token-SarBackbone=Qwen2.5-VL-7B, UQ Level=Token-level (TL)2026.05 | 0.588 | 183 | |
| Token-SarBackbone=Qwen2-1.5B, UQ Level=Token-level (TL)2026.05 | 0.587 | 130 | |
| VIG-TUQ-JSDBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.575 | 167 | |
| SEBackbone=Qwen2-1.5B, UQ Level=Semantic-level (SL)2026.05 | 0.555 | 123 | |
| Token-SarBackbone=Qwen2.5-VL-3B, UQ Level=Token-level (TL)2026.05 | 0.554 | 175 | |
| Token-SarBackbone=idefics-9B, UQ Level=Token-level (TL)2026.05 | 0.55 | 170 |