Multi-view Hallucination Evaluation on MVH-Bench Cross-Instance 1.0 (test)
67.5Binary AccuracyRSCD
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RSCDBackbone=Qwen2.5-VL-7B2026.03 | 67.5 | 45.19 | 17.69 | 73.66 | 53.7 | 257.73 | |
| VCDBackbone=Qwen2.5-VL-7B2026.03 | 67.43 | 45.79 | 19.44 | 71.71 | 50.28 | 254.65 | |
| AvisCBackbone=Qwen2.5-VL-7B2026.03 | 67.06 | 45.09 | 20 | 73.8 | 53.43 | 259.38 | |
| ICDBackbone=Qwen2.5-VL-7B2026.03 | 66.53 | 43.8 | 18.7 | 72.73 | 52.59 | 254.35 | |
| Base modelBackbone=Qwen2.5-VL-7B2026.03 | 64.1 | 41.3 | 16.29 | 66.57 | 43.52 | 231.78 | |
| RSCDBackbone=LLaVA-OneVision-7B2026.03 | 62.66 | 38.1 | 14.45 | 67.64 | 44.91 | 227.76 | |
| VCDBackbone=LLaVA-OneVision-7B2026.03 | 60.86 | 32.73 | 10.93 | 68.15 | 44.26 | 216.92 | |
| ICDBackbone=LLaVA-OneVision-7B2026.03 | 60.46 | 33.52 | 12.22 | 67.22 | 42.78 | 216.2 | |
| AvisCBackbone=LLaVA-OneVision-7B2026.03 | 60.44 | 30 | 10.19 | 67.73 | 43.15 | 211.51 | |
| Base modelBackbone=LLaVA-OneVision-7B2026.03 | 58.36 | 32.36 | 10.93 | 62.78 | 36.29 | 200.72 |