Object Hallucination on COCO 512-token budget (test)
62.9Consistency ScoreVCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VCDBackbone=mPLUG-owl2, Token Budget=5122025.12 | 62.9 | 20.4 | |
| NucleusBackbone=InstructBLIP, Token Budget=5122025.12 | 60.8 | 28.5 | |
| NucleusBackbone=mPLUG-owl2, Token Budget=5122025.12 | 60.7 | 20 | |
| VCD+CG-VLMBackbone=mPLUG-owl2, Token Budget=5122025.12 | 60.1 | 20.3 | |
| Nucleus+CG-VLMBackbone=mPLUG-owl2, Token Budget=5122025.12 | 59.2 | 19.1 | |
| Nucleus+CG-VLMBackbone=InstructBLIP, Token Budget=5122025.12 | 58.8 | 20.6 | |
| VCDBackbone=InstructBLIP, Token Budget=5122025.12 | 58.7 | 19 | |
| BeamBackbone=mPLUG-owl2, Token Budget=5122025.12 | 56.5 | 17.8 | |
| VCD+CG-VLMBackbone=InstructBLIP, Token Budget=5122025.12 | 56 | 18.6 | |
| VCD+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 55.8 | 15.5 | |
| BeamBackbone=InstructBLIP, Token Budget=5122025.12 | 55.4 | 15.9 | |
| Beam+CG-VLMBackbone=InstructBLIP, Token Budget=5122025.12 | 55.1 | 12.9 | |
| OPERABackbone=mPLUG-owl2, Token Budget=5122025.12 | 55 | 16.2 | |
| NucleusBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 54.2 | 16 | |
| VCDBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 53.9 | 15.9 | |
| Beam+CG-VLMBackbone=mPLUG-owl2, Token Budget=5122025.12 | 53.1 | 17 | |
| OPERA+CG-VLMBackbone=mPLUG-owl2, Token Budget=5122025.12 | 52.3 | 15.7 | |
| Nucleus+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 51.5 | 13.9 | |
| OPERABackbone=InstructBLIP, Token Budget=5122025.12 | 48.8 | 13.3 | |
| BeamBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 48.7 | 14 | |
| OPERA+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 46.8 | 13.4 | |
| Beam+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 46.2 | 13.2 | |
| OPERABackbone=LLaVA-v1.5 (7B), Token Budget=5122025.12 | 45.5 | 13.7 | |
| OPERA+CG-VLMBackbone=InstructBLIP, Token Budget=5122025.12 | 42 | 18.5 |