Hallucination Mitigation on SHR
22.1HSRAdaVBoost
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AdaVBoostBackbone=Qwen3-VL, Evaluation Judge=GPT-5-mini2026.02 | 22.1 | 23.3 | 3.51 | 69.84 | |
| AdaVBoostBackbone=InternVL3.5, Evaluation Judge=GPT-5-mini2026.02 | 22.4 | 25.7 | 1.63 | 26.74 | |
| VGABackbone=Qwen3-VL, Evaluation Judge=GPT-5-mini2026.02 | 22.7 | 24.3 | 3.88 | 71.64 | |
| VanillaBackbone=Qwen3-VL, Evaluation Judge=GPT-5-mini2026.02 | 23.1 | 24.7 | 3.94 | 72.62 | |
| VAFBackbone=Qwen3-VL, Evaluation Judge=GPT-5-mini2026.02 | 24 | 25.7 | 4.09 | 75.78 | |
| PAIBackbone=InternVL3.5, Evaluation Judge=GPT-5-mini2026.02 | 24.3 | 26.9 | 1.85 | 33.4 | |
| VGABackbone=InternVL3.5, Evaluation Judge=GPT-5-mini2026.02 | 25.4 | 28.7 | 2.08 | 36.22 | |
| VAFBackbone=InternVL3.5, Evaluation Judge=GPT-5-mini2026.02 | 25.9 | 29.2 | 2.13 | 37.01 | |
| VanillaBackbone=InternVL3.5, Evaluation Judge=GPT-5-mini2026.02 | 26.1 | 29.7 | 2.15 | 37.73 | |
| PAIBackbone=Qwen3-VL, Evaluation Judge=GPT-5-mini2026.02 | 26.2 | 28.5 | 3.87 | 72.21 | |
| AdaVBoostBackbone=LLaVA-NeXT, Evaluation Judge=GPT-5-mini2026.02 | 30.7 | 32.6 | 2.69 | 45.14 | |
| VGABackbone=LLaVA-NeXT, Evaluation Judge=GPT-5-mini2026.02 | 31.1 | 32.7 | 2.69 | 47.93 | |
| VanillaBackbone=LLaVA-NeXT, Evaluation Judge=GPT-5-mini2026.02 | 31.4 | 32.8 | 2.71 | 48.19 | |
| VAFBackbone=LLaVA-NeXT, Evaluation Judge=GPT-5-mini2026.02 | 31.9 | 33.5 | 2.77 | 49.3 | |
| PAIBackbone=LLaVA-NeXT, Evaluation Judge=GPT-5-mini2026.02 | 32.8 | 34.4 | 2.79 | 49.96 |