VQA Hallucination on POPE MS COCO v1 (test)
86Random AccuracyRobust mPLUG-Owl-7B
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Robust mPLUG-Owl-7BBackbone=7B, Supervised Training (SUP)=true2024.03 | 86 | — | 73 | — | 65 | — | 74.7 | — | |
| LLaVA-RLHF_13BBackbone=13B, Supervised Training (SUP)=true2024.03 | 85.2 | 38.4 | 83.9 | 38 | 82.3 | 40.5 | 83.8 | 39 | |
| LLaVA_13B M3ID + DPOBackbone=13B, Decoding Strategy=M3ID, Training=DPO2024.03 | 85.2 | 53.4 | 79.1 | 57.5 | 73.2 | 67.5 | 79.2 | 51.1 | |
| LLaVA-RLHF_7BBackbone=7B, Supervised Training (SUP)=true2024.03 | 84.8 | 39.6 | 83.3 | 41.8 | 80.7 | 44 | 82.9 | 41.8 | |
| LLaVA_13B M3IDBackbone=13B, Decoding Strategy=M3ID (Inference-time intervention)2024.03 | 84.3 | 55.6 | 77 | 61.6 | 71.3 | 68.2 | 77.5 | 61.8 | |
| LLaVA_7B M3ID + DPOBackbone=7B, Decoding Strategy=M3ID, Training=DPO2024.03 | 81.2 | 65.6 | 73.9 | 67.3 | 68.2 | 75.4 | 74.4 | 69.4 | |
| LLaVA_7B M3IDBackbone=7B, Decoding Strategy=M3ID (Inference-time intervention)2024.03 | 76 | 67.7 | 69.3 | 73.3 | 65.8 | 77.6 | 70.3 | 72.9 | |
| LLaVA_7BBackbone=7B, Decoding Strategy=Standard2024.03 | 74.8 | 75.1 | 61.8 | 86.7 | 58.1 | 90.1 | 64.9 | 84 | |
| MiniGPT4-13BBackbone=13B2024.03 | 73 | — | 67 | — | 62 | — | 74.7 | — | |
| LLaVA_13BBackbone=13B, Decoding Strategy=Standard2024.03 | 67.9 | 80.6 | 63.8 | 83.2 | 59.8 | 87.3 | 63.8 | 83.7 | |
| mPLUG-Owl-7BBackbone=7B, Supervised Training (SUP)=false2024.03 | 52 | — | 57 | — | 60 | — | 67.3 | — |