Visual Question Answering on TallyQA ZOOM-Medium
24.35GPT AccuracyMags-RL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Mags-RL2026.05 | 24.35 | 24.2 | |
| Zoom-Refine2026.05 | 18.15 | 15.8 | |
| GRIT2026.05 | 15.05 | 14.2 | |
| Direct Query2026.05 | 12.75 | 9.6 | |
| Chain-of-Thought2026.05 | 11.8 | 7.52 | |
| One-shot ICLICL shots=12026.05 | 10.2 | 8.4 |