Hallucination Evaluation on CHAIR (val)
59.5CHAIRsICD
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| ICDLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 59.5 | 15.8 | 76.3 | 103.7 | |
| VCDLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 56.3 | 16 | 77.1 | 99.6 | |
| VAFLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 51.1 | 14 | 77.9 | 96.6 | |
| ICTLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 51 | 14.9 | 76.7 | 101.2 | |
| ICDLVLM=Qwen-VL, Max new tokens=5122026.04 | 49.9 | 13.8 | 71.9 | 101.1 | |
| VanillaLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 48.8 | 13.4 | 77.7 | 98.7 | |
| NulluLVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 44.8 | 13 | 75.1 | 97.8 | |
| VTILVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 40.7 | 12 | 75.1 | 80.3 | |
| VanillaLVLM=Qwen-VL, Max new tokens=5122026.04 | 38.7 | 14.3 | 72.7 | 82.4 | |
| ICTLVLM=Qwen-VL, Max new tokens=5122026.04 | 38.1 | 11.4 | 74.4 | 85.1 | |
| VCDLVLM=Qwen-VL, Max new tokens=5122026.04 | 37.6 | 13.2 | 64.9 | 71.3 | |
| VAFLVLM=Qwen-VL, Max new tokens=5122026.04 | 36.9 | 10.6 | 72.5 | 79.5 | |
| NulluLVLM=Qwen-VL, Max new tokens=5122026.04 | 34.2 | 10.3 | 71.8 | 81 | |
| VTILVLM=Qwen-VL, Max new tokens=5122026.04 | 31.4 | 9.3 | 72.3 | 70.1 | |
| MESALVLM=LLaVA-v1.5, Max new tokens=5122026.04 | 31 | 8.6 | 75.8 | 93.5 | |
| MESALVLM=Qwen-VL, Max new tokens=5122026.04 | 29 | 8.2 | 72.7 | 83.7 |