Object Hallucination on COCO 64-token budget (test)
0.328CSVCD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VCDBackbone=InstructBLIP, Token Budget=642025.12 | 0.328 | 0.12 | |
| VCD+CG-VLMBackbone=InstructBLIP, Token Budget=642025.12 | 0.31 | 0.112 | |
| NucleusBackbone=InstructBLIP, Token Budget=642025.12 | 0.291 | 0.152 | |
| Nucleus+CG-VLMBackbone=InstructBLIP, Token Budget=642025.12 | 0.26 | 0.098 | |
| NucleusBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.26 | 0.11 | |
| VCDBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.246 | 0.079 | |
| Nucleus+CG-VLMBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.246 | 0.092 | |
| NucleusBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.245 | 0.093 | |
| VCDBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.241 | 0.094 | |
| BeamBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.217 | 0.075 | |
| BeamBackbone=InstructBLIP, Token Budget=642025.12 | 0.213 | 0.074 | |
| OPERABackbone=mPLUG-owl2, Token Budget=642025.12 | 0.211 | 0.079 | |
| Beam+CG-VLMBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.21 | 0.074 | |
| VCD+CG-VLMBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.21 | 0.081 | |
| Beam+CG-VLMBackbone=InstructBLIP, Token Budget=642025.12 | 0.208 | 0.062 | |
| VCD+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.208 | 0.07 | |
| OPERABackbone=InstructBLIP, Token Budget=642025.12 | 0.201 | 0.067 | |
| OPERA+CG-VLMBackbone=mPLUG-owl2, Token Budget=642025.12 | 0.198 | 0.072 | |
| Nucleus+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.196 | 0.065 | |
| BeamBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.195 | 0.061 | |
| OPERABackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.192 | 0.065 | |
| OPERA+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.188 | 0.061 | |
| OPERA+CG-VLMBackbone=InstructBLIP, Token Budget=642025.12 | 0.182 | 0.077 | |
| Beam+CG-VLMBackbone=LLaVA-v1.5 (7B), Token Budget=642025.12 | 0.176 | 0.058 |