Object Presence Hallucination Evaluation on POPE A-OKVQA 2022 (Adversarial)
80.13AccuracyVCD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| VCDModel=Qwen-VL, Decoding=VCD2026.01 | 80.13 | 79.35 | 81.47 | 80.39 | |
| SDCDModel=Qwen-VL, Decoding=SDCD2026.01 | 79.63 | 78.29 | 82 | 80.1 | |
| Regular DecodingModel=Qwen-VL, Decoding=Regular2026.01 | 79.27 | 79.99 | 78.07 | 79.01 | |
| AVISCModel=InstructBLIP2024.05 | 72.53 | 67.12 | 88.33 | 76.28 | |
| VCDModel=InstructBLIP2024.05 | 69.7 | 64.54 | 87.47 | 74.27 | |
| M3IDModel=InstructBLIP2024.05 | 69.57 | 64.21 | 88.4 | 74.39 | |
| AVISCModel=LLaVA-1.52024.05 | 68.97 | 62.7 | 93.67 | 75.11 | |
| baseModel=InstructBLIP2024.05 | 68.8 | 63.57 | 88.07 | 73.84 | |
| M3IDModel=LLaVA-1.52024.05 | 68.1 | 61.99 | 93.6 | 74.58 | |
| baseModel=LLaVA-1.52024.05 | 67.9 | 62.11 | 91.8 | 74.09 | |
| VCDModel=LLaVA-1.52024.05 | 67.43 | 61.5 | 93.2 | 74.11 |