Hallucination Evaluation on POPE Random
89.87AccuracyCAST
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CASTBackbone=LLaVA-1.5-7B2026.05 | 89.87 | — | |
| ESCModel=LLaVA [50]2026.07 | 89.57 | 88.73 | |
| GAP (LH+PCA+DA)Latent Head=true, PCA=true, Data Augmentation=true2026.05 | 88.7 | — | |
| BaselineModel=LLaVA [50]2026.07 | 87.9 | 86.65 | |
| LH+DA (no PCA)Latent Head=true, Data Augmentation=true, PCA=false2026.05 | 87.09 | — | |
| Qwen2.5-VL-7BModel size=7B2026.05 | 87.03 | — | |
| HALVABackbone=LLaVA-1.5-7B2026.05 | 86.4 | — | |
| HADPOBackbone=LLaVA-1.5-7B2026.05 | 86 | — | |
| LLaVA-1.5-7B2026.05 | 83.29 | — | |
| ESCModel=Qwen2 [92]2026.07 | 81.27 | 77.04 | |
| BaselineModel=Qwen2 [92]2026.07 | 50.83 | 3.41 |