Truthfulness and Calibration Evaluation on Cross-LVLM Pooled Average (GQA, POPE, etc.)
7.1ECEBICR
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| BICRAggregation=Cross-LVLM Pooled Average2026.05 | 7.1 | 18.4 | 71.5 | 76.9 | 87.5 | 78.6 | |
| IIAggregation=Cross-LVLM Pooled Average2026.05 | 8.3 | 19.9 | 69 | 78.4 | 84.3 | 74.8 | |
| PIKAggregation=Cross-LVLM Pooled Average2026.05 | 9.3 | 19.4 | 69.8 | 78.3 | 86.3 | 76.6 | |
| SAPLMAAggregation=Cross-LVLM Pooled Average2026.05 | 12.2 | 21.3 | 69.5 | 79 | 81.6 | 72.9 | |
| CCPSAggregation=Cross-LVLM Pooled Average2026.05 | 15.3 | 27.1 | 63.8 | 72.1 | 72.8 | 63.1 | |
| PEAggregation=Cross-LVLM Pooled Average2026.05 | 18.5 | 26.1 | 63 | 77.1 | 69.3 | 60 | |
| Self-ProbingAggregation=Cross-LVLM Pooled Average2026.05 | 27.5 | 29.8 | 64.5 | 77.5 | 77.8 | 65.6 | |
| P(True)Aggregation=Cross-LVLM Pooled Average2026.05 | 38 | 39.8 | 52.1 | 58.6 | 72.8 | 55.7 |