Differences Caption Generation on MagicBrush (test)
50Main Difference CountGPT-4o
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GPT-4o2025.06 | 50 | 11 | 12 | 61 | 57 | 1.9 | 0 | |
| Qwen2.5 VL2025.06 | 37 | 11 | 1 | 63 | 59 | 1.5 | 60 | |
| GPT-42025.06 | 34 | 8 | 8 | 75 | 74 | 2.5 | 80 | |
| GPT-4 Turbo2025.06 | 34 | 7 | 8 | 78 | 75 | 1.5 | 450 | |
| InternVL32025.06 | 25 | 8 | 11 | 88 | 83 | 3.4 | 230 | |
| LLaVAtraining=Supervised2025.06 | 12 | — | — | — | — | — | — | |
| LLaVA2025.06 | 5 | — | — | — | — | — | — |