Text-to-CAD Generation on Multi-part CAD
79.06COVTOOLCAD
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| TOOLCADFine-Tuning=True (+GRPO), VLM-Based=False, Backbone=Qwen2.5-7B-Instruct, Prompt Type=L3 text, CAD-Code=False2026.04 | 79.06 | 80.63 | 1.36 | 3.21 | |
| CAD-LlamaFine-Tuning=True, VLM-Based=CLIP, Backbone=LLaMA3-8B-Instruct, Prompt Type=L0~L1 text, CAD-Code=True2026.04 | 74.86 | 79.46 | 1.62 | 3.85 | |
| CADFusionFine-Tuning=True (+DPO), VLM-Based=GPT-4o, Backbone=LLaMA3-8B-Instruct, Prompt Type=L1~L2 text, CAD-Code=False2026.04 | 71.93 | 72.65 | 3.07 | 5.03 | |
| FlexCADFine-Tuning=True, VLM-Based=False, Backbone=LLaMA3-8B-Instruct, Prompt Type=L3 text token, CAD-Code=False2026.04 | 70.49 | 81.43 | 2.35 | 4.79 | |
| CAD-CoderFine-Tuning=True (+GRPO), VLM-Based=False, Backbone=Qwen2.5-7B-Instruct, Prompt Type=L0~L3 text, CAD-Code=True2026.04 | 65.48 | 57.34 | 5.73 | 11.42 | |
| CAD TranslatorFine-Tuning=True, VLM-Based=False, Backbone=Transformer, Prompt Type=L0 text, CAD-Code=False2026.04 | 65.22 | 69.28 | 3.79 | 9.14 | |
| CAD-GPTFine-Tuning=True, VLM-Based=LLaVA, Backbone=LLaVA, Prompt Type=L0 text + image, CAD-Code=False2026.04 | 64.59 | 52.78 | 4.49 | 22.17 | |
| CADCodeVerifyFine-Tuning=False, VLM-Based=GPT-4/Gemini-1.5-Pro, Backbone=GPT-4/Gemini-1.5-Pro/CodeLlama, Prompt Type=L2~L3 text, CAD-Code=True2026.04 | 64.37 | — | 5.46 | 18.26 | |
| Text2CADFine-Tuning=True, VLM-Based=False, Backbone=BERT, Prompt Type=L0~L3 text, CAD-Code=False2026.04 | 61.94 | 60.39 | 5.19 | 10.28 | |
| CAD-AssisantFine-Tuning=False, VLM-Based=GPT-4o, Backbone=GPT-4o, Prompt Type=L0 text + docstr + image, CAD-Code=True2026.04 | 60.32 | 51.25 | 6.13 | 25.09 | |
| RLCADFine-Tuning=True (+PPO), VLM-Based=False, Backbone=Transformer, Prompt Type=/, CAD-Code=False2026.04 | 55.18 | — | 7.45 | 8.33 | |
| VideoCADFine-Tuning=True, VLM-Based=DINOv2/CLIP, Backbone=ViT, Prompt Type=video frame, CAD-Code=False2026.04 | 49.25 | — | 11.75 | 32.19 |