Failure Reasoning and Correction on Dream2Fix (test)
91.3ROUGE-LDream2Fix-VLM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Dream2Fix-VLM2026.03 | 91.3 | 94.1 | 95 | 77.1 | 81.3 | |
| GPT-4o2026.03 | 50.2 | 53.8 | 62.3 | 61.5 | 19.7 | |
| Gemini-1.5-Flash2026.03 | 40.6 | 51.8 | 67.9 | 49.6 | 17.9 | |
| Qwen2-VL-72B2026.03 | 24.8 | 38.6 | 56.7 | 24.8 | 14.3 | |
| Qwen2.5-VL-7B2026.03 | 17 | 39.1 | 55.1 | 24.1 | 11.3 | |
| LLaVA-NeXT-34B2026.03 | 13 | 12.7 | 15.3 | 45.3 | 4.4 | |
| LLaVA-NeXT-7B2026.03 | 8 | 9.9 | 11.1 | 32.2 | 2.2 |