Categorical Visual Question Answering on POPE
96.1AccuracySN
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SNBase Model=Qwen-VL-4b-Instruct, Probing Set Size=3,000, Configuration=Best probed2026.03 | 96.1 | 95.9 | |
| Qwen-VL-4b-InstructProbing Set Size=3,0002026.03 | 95 | 94.8 | |
| SNBase Model=LLaVA-v1.5-7b, Probing Set Size=3,000, Configuration=Best probed2026.03 | 92.5 | 92.3 | |
| LLaVA-v1.5-7bProbing Set Size=3,0002026.03 | 90.7 | 91.1 |