Visual Question Answering on POPE (val)
93.7AccuracySNs (maj. voting)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| SNs (maj. voting)Backbone=Qwen3-VL-4b-Instruct2026.03 | 93.7 | 95.8 | 96.5 | 91.1 | 93.7 | |
| SNs (mean)Backbone=Qwen3-VL-4b-Instruct2026.03 | 93.4 | 95.8 | 97.6 | 89.3 | 93.3 | |
| Qwen3-VL-4b-Instruct2026.03 | 91.5 | — | 98.5 | 84.9 | 91.2 | |
| SNs (mean)Backbone=LLaVA-v1.5-7b2026.03 | 90.9 | 92 | 92.6 | 89.5 | 91 | |
| SNs (maj. voting)Backbone=LLaVA-v1.5-7b2026.03 | 90.9 | 92 | 91 | 91.4 | 91.2 | |
| LLaVA-v1.5-7b2026.03 | 89.8 | — | 89.5 | 90.8 | 90.2 |