Hallucination Evaluation on Object HalBench (full benchmark)
25.2Ha (Living Room)LLaVA-1.5
Evaluation results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| LLaVA-1.52023.12 | 25.2 | 41.8 | 16.6 | 18.9 | 23.9 | 5 | 22.4 | 30.4 | 8 | 20.6 | 28 | 7.4 | 9.2 | |
| QWEN-VL2023.12 | 24.5 | 34.5 | 10 | 16.4 | 20.8 | 4.4 | 21.6 | 17.5 | -4.1 | 22.5 | 32 | 9.5 | 5 | |
| LLaVA-RLHF2023.12 | 23.7 | 34.5 | 10.8 | 13.1 | 17.4 | 4.3 | 18.2 | 19.5 | 1.4 | 18.3 | 22.7 | 4.4 | 5.2 | |
| GPT-4V2023.12 | 8.2 | 19.4 | 11.2 | 4.6 | 5.7 | 1.1 | 5.9 | 13.3 | 7.5 | 4.2 | 4.6 | 0.4 | 5 | |
| RLHF-V2023.12 | 5.5 | 8 | 2.5 | 3.8 | 5.9 | 2.1 | 4.1 | 4 | -0.1 | 2.3 | 4.6 | 2.3 | 1.7 |