Object Hallucination Evaluation on POPE (test) with Efficiency Metrics
26Total TimeDeSAP
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| DeSAPBase Model=LLaVA-1.5-7B, Token Pruning Strategy=DeSAP, Pruning Ratio=11.1%2026.04 | 26 | 0.22 | 0.42 | 85.8 | |
| HoloVBase Model=LLaVA-1.5-7B, Token Pruning Strategy=HoloV, Pruning Ratio=11.1%2026.04 | 29 | 0.35 | 0.42 | 80.3 | |
| SparseVLMBase Model=LLaVA-1.5-7B, Token Pruning Strategy=SparseVLM, Pruning Ratio=11.1%2026.04 | 35 | 0.43 | 1.31 | 75.1 | |
| LLaVA-1.5-7BBase Model=LLaVA-1.5-7B, Token Pruning Strategy=None, Pruning Ratio=0%2026.04 | 42 | 0.52 | 3.82 | 85.9 |