Difference Caption Generation on EditInspector 1.0 (test)
39Main Difference CountGPT-4o
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| GPT-4omode=zero-shot2025.06 | 39 | 12 | 14 | 60 | 56 | 1.8 | 0.3 | |
| Qwen2.5-VLmode=zero-shot2025.06 | 38 | 12 | 14 | 58 | 52 | 1.5 | 0.7 | |
| Geminimode=zero-shot2025.06 | 31 | — | — | — | — | — | — | |
| Gemini-1.5mode=zero-shot2025.06 | 31 | 8 | 9 | 67 | 65 | 1 | 24 | |
| GPT-4mode=zero-shot2025.06 | 27 | 8 | 10 | 78 | 75 | 2.5 | 0.7 | |
| InternVL2mode=zero-shot2025.06 | 26 | 7 | 10 | 87 | 83 | 3.1 | 3.4 | |
| GPT-4 Turbomode=zero-shot2025.06 | 24 | 8 | 9 | 75 | 72 | 1.5 | 6 | |
| LLaVASupervised=true2025.06 | 10 | — | — | — | — | — | — | |
| LLaVAmode=zero-shot2025.06 | 8 | — | — | — | — | — | — |