Object Hallucination Mitigation on COCO 2014
56Cs ScoreNucleus
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NucleusBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 56 | 16.4 | 73.4 | 71 | 72.2 | |
| VCDBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 55.4 | 15.7 | 75.6 | 71.8 | 73.7 | |
| NucleusBase Model=Average, Max Length=5122026.03 | 55 | 15.7 | — | — | — | |
| NucleusBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 54 | 15 | 74.6 | 73 | 73.8 | |
| VCDBase Model=Average, Max Length=5122026.03 | 53.8 | 14.8 | — | — | — | |
| OPERABase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 52.6 | 14.1 | 68.1 | 73.1 | 70.5 | |
| VCDBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 52.3 | 13.9 | 75.5 | 74.6 | 75.1 | |
| BeamBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 52.1 | 14.1 | 78.5 | 75.2 | 76.8 | |
| BeamBase Model=Average, Max Length=5122026.03 | 50.9 | 13.4 | — | — | — | |
| OPERABase Model=Average, Max Length=5122026.03 | 50.6 | 13.9 | — | — | — | |
| BeamBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 49.7 | 12.7 | 79.1 | 77.3 | 78.2 | |
| OPERABase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 48.5 | 13.7 | 78.1 | 75.5 | 76.8 | |
| GreedyBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 47.9 | 13.6 | 78.1 | 75.8 | 77 | |
| GreedyBase Model=Average, Max Length=5122026.03 | 46.1 | 12.6 | — | — | — | |
| GreedyBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 44.3 | 11.7 | 77.6 | 78.7 | 78.2 | |
| PAIBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 30.1 | 8.8 | 70.8 | 83.7 | 76.7 | |
| PAIBase Model=Average, Max Length=5122026.03 | 29.5 | 8.6 | — | — | — | |
| PAIBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 28.9 | 8.5 | 70.4 | 83.4 | 76.4 | |
| VARBase Model=LLaVA-1.5-13B, Max Length=5122026.03 | 26.7 | 8.8 | 67.3 | 84.1 | 74.7 | |
| VARBase Model=Average, Max Length=5122026.03 | 26 | 7.6 | — | — | — | |
| VARBase Model=LLaVA-1.5-7B, Max Length=5122026.03 | 25.3 | 6.5 | 69.2 | 86.6 | 76.9 | |
| SAE (Ours)Base Model=LLaVA-1.5-7B, Max Length=5122026.03 | 24.2 | 5.9 | 68.2 | 87.3 | 76.6 | |
| SAE (Ours)Base Model=Average, Max Length=5122026.03 | 24 | 7.2 | — | — | — | |
| SAE (Ours)Base Model=LLaVA-1.5-13B, Max Length=5122026.03 | 23.8 | 8.5 | 64.8 | 85.9 | 73.9 |