Object Hallucination Discrimination on POPE-GQA Adversarial
81.87AccuracyCMPC
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CMPCBase Model=LLaVA 1.52025.01 | 81.87 | 79.18 | 86.47 | 82.66 | |
| ICDBase Model=LLaVA 1.52025.01 | 77.6 | 76.04 | 80.6 | 78.25 | |
| CMPCBase Model=InstructBLIP2025.01 | 76.83 | 73.94 | 82.87 | 78.15 | |
| VCDBase Model=LLaVA 1.52025.01 | 76.09 | 70.83 | 88.75 | 78.78 | |
| Instruction Contrastive Decoding (ICD)Backbone=InstructBLIP2024.03 | 75.17 | 70.59 | 86.27 | 77.65 | |
| ICDBase Model=InstructBLIP2025.01 | 75.17 | 70.59 | 86.27 | 77.65 | |
| Visual Contrastive Decoding (VCD)Backbone=InstructBLIP2024.03 | 75.08 | 70.59 | 85.99 | 77.53 | |
| BaselineBase Model=LLaVA 1.52025.01 | 75.08 | 73.19 | 79.16 | 76.06 | |
| VCDBase Model=InstructBLIP2025.01 | 75.08 | 70.59 | 85.99 | 77.53 | |
| Standard DecodingBackbone=InstructBLIP2024.03 | 70.56 | 66.12 | 84.33 | 74.12 | |
| BaselineBase Model=InstructBLIP2025.01 | 70.56 | 66.12 | 84.33 | 74.12 | |
| Instruction Contrastive Decoding (ICD)Backbone=miniGPT42024.03 | 59.64 | 58.21 | 76.81 | 66.23 | |
| Visual Contrastive Decoding (VCD)Backbone=miniGPT42024.03 | 57.78 | 57.7 | 69.82 | 63.18 | |
| Standard DecodingBackbone=miniGPT42024.03 | 56.75 | 56.26 | 67.99 | 61.57 |