Visual Hallucination Evaluation on AMBER Generative
6.3CHAIRiVicuna-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Vicuna-7BBackbone=Vicuna-7B2026.06 | 6.3 | 52.2 | 30.8 | 3.3 | |
| HeRABackbone=Vicuna-7B2026.06 | 6.1 | 52.6 | 31.9 | 3.5 | |
| Vicuna-13BBackbone=Vicuna-13B2026.06 | 6.1 | 52.5 | 28.4 | 3.3 | |
| HeRABackbone=Vicuna-13B2026.06 | 6.1 | 52.9 | 30.3 | 3.2 | |
| Qwen2.5-14BBackbone=Qwen2.5-14B2026.06 | 6.1 | 52.3 | 29.4 | 3.6 | |
| Qwen2.5-3BBackbone=Qwen2.5-3B2026.06 | 5.8 | 52 | 30.1 | 3.3 | |
| Qwen3-8BBackbone=Qwen3-8B2026.06 | 5.8 | 52.8 | 28.9 | 3.2 | |
| HeRABackbone=LLama3-8B2026.06 | 5.6 | 51.7 | 26.6 | 2.7 | |
| Qwen3-4BBackbone=Qwen3-4B2026.06 | 5.5 | 53 | 28.2 | 3 | |
| HeRABackbone=Qwen3-4B2026.06 | 5.5 | 52.5 | 27.6 | 2.8 | |
| LLama3-8BBackbone=LLama3-8B2026.06 | 5.5 | 51.5 | 27.5 | 2.9 | |
| Qwen3-14BBackbone=Qwen3-14B2026.06 | 5.5 | 53.1 | 28.8 | 3.2 | |
| HeRABackbone=Qwen2.5-14B2026.06 | 5.4 | 51.9 | 28 | 3.1 | |
| Qwen2.5-7BBackbone=Qwen2.5-7B2026.06 | 5.3 | 52.3 | 26.7 | 2.7 | |
| HeRABackbone=Qwen2.5-3B2026.06 | 5.2 | 52.3 | 27.6 | 2.8 | |
| HeRABackbone=Qwen3-8B2026.06 | 5.2 | 53.1 | 27.9 | 3 | |
| HeRABackbone=Qwen3-14B2026.06 | 5 | 53 | 26.7 | 3 | |
| HeRABackbone=Qwen2.5-7B2026.06 | 4.9 | 52.5 | 25.1 | 2.7 |