Metric Sensitivity Analysis on Quilt-1M Visual Hallucination
0.9Performance ScoreBERTScore
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BERTScoreFocus=Semantic2026.03 | 0.9 | 2.2 | |
| PathGLS (Sl)Focus=Consistency2026.03 | 0.82 | 9.9 | |
| PathGLS (Sg)Focus=Visual-Text2026.03 | 0.46 | 40.3 | |
| RadGraphFocus=Entity2026.03 | 0.19 | 38.7 | |
| BLEU-4Focus=Lexical2026.03 | 0.12 | 25 |