Visual Illusion and Hallucination Evaluation on HallusionBench (HallB)
41.7HallB ScoreQwen2-VL-2B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2-VL-2Barchitecture=Compositional, params=2.2B2024.12 | 41.7 | |
| HoVLEarchitecture=Monolithic, params=2.6B2024.12 | 39.6 | |
| HoVLEarchitecture=Monolithic, params=2.6B, resolution=HD2024.12 | 38.4 | |
| InternVL2architecture=Compositional, params=2.2B2024.12 | 37.9 | |
| InternVL1.5architecture=Compositional, params=2.2B2024.12 | 37.5 | |
| MiniCPM-Varchitecture=Compositional, params=2.8B2024.12 | 36.2 | |
| MiniCPM-V-2architecture=Compositional, params=2.8B2024.12 | 36.1 | |
| PaliGemma-3Barchitecture=Compositional, params=2.9B2024.12 | 32.2 | |
| VILA1.5-3Barchitecture=Compositional, params=3.1B2024.12 | 31.2 | |
| DeepSeek-VL-1.3Barchitecture=Compositional, params=2.0B2024.12 | 27.6 | |
| EVE-7Barchitecture=Monolithic, params=7B, resolution=HD2024.12 | 26.4 | |
| EVE-7Barchitecture=Monolithic, params=7B2024.12 | 21.1 | |
| Chameleon-7Barchitecture=Monolithic, params=7B2024.12 | 17.1 |