Science Question Answering on ScienceQA (Image Acc.)
88.6Image AccuracyQwen2.5-VL-7B-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7B-InstructMitigation=None (baseline)2025.07 | 88.6 | |
| SENTINELBase Model=Qwen2.5-VL-7B-Instruct2025.07 | 88.5 | |
| SENTINELBase Model=Qwen2-VL-7B-Instruct2025.07 | 86.9 | |
| Qwen2-VL-7B-InstructMitigation=None (baseline)2025.07 | 85.7 | |
| SENTINELBase Model=LLaVA-v1.6-vicuna-13B2025.07 | 78 | |
| LLaVA-v1.6-vicuna-13BMitigation=None (baseline)2025.07 | 77.7 | |
| SENTINELBase Model=Qwen2-VL-2B-Instruct2025.07 | 77.4 | |
| Qwen2-VL-2B-InstructMitigation=None (baseline)2025.07 | 76.9 | |
| LLaVA-v1.6-vicuna-7BMitigation=None (baseline)2025.07 | 74.3 | |
| SENTINELBase Model=LLaVA-v1.6-vicuna-7B2025.07 | 74.2 | |
| SENTINELModel=LLaVA-v1.5-13B2025.07 | 72.8 | |
| VCDModel=LLaVA-v1.5-13B2025.07 | 72 | |
| vanilla-DPOModel=LLaVA-v1.5-13B2025.07 | 71.8 | |
| baselineModel=LLaVA-v1.5-13B2025.07 | 71.6 | |
| HSA-DPOModel=LLaVA-v1.5-13B2025.07 | 71.3 | |
| HA-DPOModel=LLaVA-v1.5-7B2025.07 | 69.7 | |
| SENTINELModel=LLaVA-v1.5-7B2025.07 | 69.2 | |
| POVIDModel=LLaVA-v1.5-7B2025.07 | 68.8 | |
| VCDModel=LLaVA-v1.5-7B2025.07 | 68.7 | |
| OPERAModel=LLaVA-v1.5-7B2025.07 | 68.2 | |
| RLAIF-VModel=LLaVA-v1.5-7B2025.07 | 68.2 | |
| CLIP-DPOModel=LLaVA-v1.5-7B2025.07 | 67.6 | |
| DoLaModel=LLaVA-v1.5-7B2025.07 | 67.5 | |
| TPOModel=LLaVA-v1.5-7B2025.07 | 67.1 | |
| baselineModel=LLaVA-v1.5-7B2025.07 | 66.8 | |
| EFUFModel=LLaVA-v1.5-7B2025.07 | 66.4 |