Hallucination Mitigation on SHR (test)
5.36SPIDoLa(high)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DoLa(high)Base Model=LLaVA-1.5, M-level=high2025.04 | 5.36 | 95.69 | 2.36 | 44.55 | 44 | 47 | |
| Beam3Base Model=LLaVA-1.5, Beam Size=32025.04 | 5.15 | 93.96 | 2.2 | 42.97 | 43 | 46 | |
| Beam5Base Model=LLaVA-1.5, Beam Size=52025.04 | 5.12 | 93.81 | 2.26 | 44.33 | 44 | 47 | |
| GreedyBase Model=LLaVA-1.5, Decoding Strategy=Greedy2025.04 | 5.08 | 89.73 | 2.1 | 39.76 | 42 | 45 | |
| VCDBase Model=LLaVA-1.52025.04 | 5.08 | 90.54 | 2.4 | 45.44 | 48 | 51 | |
| DoLa(low)Base Model=LLaVA-1.5, M-level=low2025.04 | 5.02 | 89.92 | 2.04 | 38.79 | 41 | 43 | |
| AGLABase Model=LLaVA-1.52025.04 | 4.95 | 88.42 | 2.11 | 39.96 | 43 | 46 | |
| TARACBase Model=LLaVA-1.5, alpha=0.3, beta=0.92025.04 | 4.93 | 82.75 | 1.64 | 28.57 | 33 | 35 | |
| OPERABase Model=LLaVA-1.52025.04 | 4.79 | 85.67 | 1.95 | 37.18 | 41 | 44 |