Saliency-grounded Question Answering on SalBench p3 (test)
89.2AccuracyGPT-4o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4o2026.04 | 89.2 | |
| Claude-3.5 Sonnet2026.04 | 86.7 | |
| Qwen2.5-VL-7B + Saliency-R1Training Stage=Saliency-R12026.04 | 63.7 | |
| Qwen2.5-VL-3B + Saliency-R1Training Stage=Saliency-R12026.04 | 59.6 | |
| Qwen2.5-VL-7B + SFTTraining Stage=SFT2026.04 | 57.1 | |
| Qwen2.5-VL-3B + SFTTraining Stage=SFT2026.04 | 51.7 | |
| Qwen2.5-VL-7BTraining Stage=Base2026.04 | 49.1 | |
| Qwen2.5-VL-3BTraining Stage=Base2026.04 | 40.4 |