Selective Prediction on POPE 60 (test)
1AUROCRidge LTSA Chart
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Ridge LTSA ChartModel=LLaVA-v1.5-7B, Variant=TRiE/Ridge Proximity2026.06 | 1 | 0.044 | 0.953 | 0.04 | |
| logP(x)Model=LLaVA-v1.5-7B, Variant=Mean logP(x) per token2026.06 | 0.817 | 0.106 | 0.891 | 0.077 | |
| Semantic EntropyModel=LLaVA-v1.5-7B2026.06 | 0.74 | 0.133 | 0.864 | 0.093 |