Object Hallucination Evaluation on POPE (F1 Scores)
86.93Random F1 ScoreTPC
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| TPCIntegration Strategy=Ours (core–residual alignment), Base LVLM=LLaVA-1.5-7B2026.07 | 86.93 | 85.81 | 82.74 | 85.16 | |
| Adv-W2SIntegration Strategy=Robust vision encoder replacement, Base LVLM=LLaVA-1.5-7B2026.07 | 85.6 | 84.9 | 81 | 83.8 | |
| VL-SAEIntegration Strategy=Decoding / concept-level enhancement (frozen LVLM), Base LVLM=LLaVA-1.5-7B2026.07 | 85.5 | 84.37 | 82.29 | 84.05 | |
| VCDIntegration Strategy=Decoding / concept-level enhancement (frozen LVLM), Base LVLM=LLaVA-1.5-7B2026.07 | 84.04 | 82.31 | 80.13 | 82.16 | |
| FAREIntegration Strategy=Robust vision encoder replacement, Base LVLM=LLaVA-1.5-7B2026.07 | 82.2 | 81.5 | 78.6 | 80.8 | |
| PMG-AFTIntegration Strategy=Robust vision encoder replacement, Base LVLM=LLaVA-1.5-7B2026.07 | 81.7 | 80.9 | 76.3 | 79.6 | |
| Patch-Aligned TrainingIntegration Strategy=Projector / fine-grained alignment (LVLM-side training), Base LVLM=LLaVA-1.5-7B2026.07 | 81.2 | 80.5 | 78.1 | 79.93 | |
| LLaVA1.5 (Regular)Integration Strategy=Decoding / concept-level enhancement (frozen LVLM), Base LVLM=LLaVA-1.5-7B2026.07 | 80.87 | 79.27 | 77.16 | 79.1 | |
| TGA-ZSRIntegration Strategy=Robust vision encoder replacement, Base LVLM=LLaVA-1.5-7B2026.07 | 80.4 | 79.8 | 76 | 78.7 | |
| TeCoAIntegration Strategy=Robust vision encoder replacement, Base LVLM=LLaVA-1.5-7B2026.07 | 79.8 | 79.1 | 75.2 | 78 |