Program Error Correction on CADReview Human-made programs 1.0 (test)
33.2RLReCAD-QW
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ReCAD-QWbackbone=QWen2-VL, fine-tuning=true2025.05 | 33.2 | 67.47 | 63.6 | 4.42 | 0.52 | 30.87 | 10.59 | |
| ReCAD-LAbackbone=LLaVA-OV2025.05 | 33.14 | 67.43 | 59.15 | 4.76 | 0.75 | 33.09 | 13.74 | |
| Llama 3.2fine-tuning=true2025.05 | 32.65 | 67.23 | 51.24 | 5.82 | 1.67 | 36.18 | 16.14 | |
| GPT-4o2025.05 | 29.97 | 64.81 | 31.84 | 5.52 | 1.55 | 48.7 | 18.57 | |
| Gemini 2.02025.05 | 25.61 | 59.95 | 23.36 | 5.97 | 1.6 | 55.85 | 34.88 | |
| Claude 3.52025.05 | 24.88 | 60.23 | 21.51 | 9.03 | 2.63 | 79.63 | 86.09 |