Object Hallucination Evaluation on MSCOCO (val)
57CHAIRSVCD
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| VCDModel=LLaVA-1.5, Decoding strategy=nucleus sampling2026.01 | 57 | 16.9 | 74.4 | — | |
| RegularModel=LLaVA-1.5, Decoding strategy=nucleus sampling2026.01 | 55.6 | 17.3 | 72.2 | — | |
| PAIModel=LLaVA-1.5, Decoding strategy=nucleus sampling2026.01 | 38.2 | 11 | 73.4 | — | |
| VanillaAvg. FLOPs(T)=34.322026.03 | 26.1 | 7.8 | — | — | |
| PinPointAvg. FLOPs(T)=22.182026.03 | 25.6 | 7.1 | — | — | |
| SDCDModel=LLaVA-1.5, Decoding strategy=nucleus sampling2026.01 | 18.6 | 6.4 | 70.7 | — | |
| BaselineBase Model=LLaVA 1.5, Decoding Strategy=Sampling2025.01 | 17.8 | 55.6 | 77 | 72.4 | |
| BaselineBase Model=InstructBLIP, Decoding Strategy=Sampling2025.01 | 16.7 | 57 | 76.3 | 70.4 | |
| VCDBase Model=LLaVA 1.5, Decoding Strategy=Sampling2025.01 | 16.4 | 54.2 | 80 | 76.7 | |
| VCDBase Model=LLaVA 1.5, Decoding Strategy=Greedy2025.01 | 15.7 | 55.2 | 80.8 | 77.6 | |
| VCDBase Model=InstructBLIP, Decoding Strategy=Sampling2025.01 | 15.6 | 53.6 | 79.2 | 74.6 | |
| ICDBase Model=LLaVA 1.5, Decoding Strategy=Sampling2025.01 | 15.4 | 54 | 80.7 | 77.2 | |
| ICDBase Model=LLaVA 1.5, Decoding Strategy=Greedy2025.01 | 15.4 | 57 | 80.9 | 77.5 | |
| BaselineBase Model=LLaVA 1.5, Decoding Strategy=Greedy2025.01 | 15.1 | 49.6 | 81.2 | 77.9 | |
| ICDBase Model=InstructBLIP, Decoding Strategy=Sampling2025.01 | 14.5 | 49.6 | 78.5 | 72.6 | |
| VCDBase Model=InstructBLIP, Decoding Strategy=Greedy2025.01 | 14 | 49.8 | 80.1 | 74.9 | |
| BaselineBase Model=InstructBLIP, Decoding Strategy=Greedy2025.01 | 13.3 | 48.2 | 79.7 | 73.8 | |
| CMACBase Model=InstructBLIP, Decoding Strategy=Sampling2025.01 | 13.2 | 50.6 | 79.7 | 73.7 | |
| CMACBase Model=LLaVA 1.5, Decoding Strategy=Sampling2025.01 | 12.7 | 47 | 81 | 75.6 | |
| ICDBase Model=InstructBLIP, Decoding Strategy=Greedy2025.01 | 12.6 | 44.6 | 80.1 | 73.9 | |
| CMACBase Model=LLaVA 1.5, Decoding Strategy=Greedy2025.01 | 12.5 | 47.6 | 81.5 | 76.2 | |
| VCDBase Model=Qwen-VL, Decoding Strategy=Greedy2025.01 | 12.2 | 46.6 | 81.7 | 76.3 | |
| VCDBase Model=Qwen-VL, Decoding Strategy=Sampling2025.01 | 12.1 | 47.2 | 81.6 | 76.1 | |
| CMACBase Model=InstructBLIP, Decoding Strategy=Greedy2025.01 | 12.1 | 45.2 | 80.5 | 74.2 | |
| BaselineBase Model=Qwen-VL, Decoding Strategy=Greedy2025.01 | 11.4 | 45.8 | 81 | 74.6 | |
| BaselineBase Model=Qwen-VL, Decoding Strategy=Sampling2025.01 | 11.3 | 44.8 | 81.1 | 74.6 | |
| ICDBase Model=Qwen-VL, Decoding Strategy=Sampling2025.01 | 11.2 | 43.2 | 81.5 | 75 | |
| ICDBase Model=Qwen-VL, Decoding Strategy=Greedy2025.01 | 11.1 | 42.6 | 81.3 | 74.9 | |
| CMACBase Model=Qwen-VL, Decoding Strategy=Greedy2025.01 | 10.9 | 41.4 | 81.8 | 75.6 | |
| CMACBase Model=Qwen-VL, Decoding Strategy=Sampling2025.01 | 10.6 | 41.2 | 81.8 | 75.4 |