Img2CAD: Reverse Engineering 3D CAD Models from Images through VLM-Assisted Conditional Factorization

Published in SIGGRAPH Asia, 2025

Yang You, Mikaela Angelina Uy, Jiaqi Han, Rahul Thomas, Haotong Zhang, Yi Du, Hansheng Chen, Francis Engelmann, Suya You, Leonidas Guibas

arXiv / PDF / code / project page / BibTeX

Abstract

Img2CAD reverse engineers editable 3D CAD programs from single images. A finetuned vision-language model (Llama 3.2) predicts the discrete program structure with semantic part labels, and TrAssembler, a transformer flow-matching model with symmetry guidance, predicts the continuous attributes. We also release a dataset of 4,574 ShapeNet chairs, tables and cabinets turned into sketch-and-extrude CAD programs.

BibTeX

@inproceedings{you2025img2cad,
  title     = {Img2CAD: Reverse Engineering 3D CAD Models from Images through VLM-Assisted Conditional Factorization},
  author    = {Yang You and Mikaela Angelina Uy and Jiaqi Han and Rahul Thomas and Haotong Zhang and Yi Du and Hansheng Chen and Francis Engelmann and Suya You and Leonidas Guibas},
  booktitle = {ACM SIGGRAPH Asia},
  year      = {2025},
  eprint    = {2408.01437},
  archivePrefix = {arXiv}
}

Recommended citation: You, Y., Uy, M.A., Han, J., Thomas, R., Zhang, H., Du, Y., Chen, H., Engelmann, F., You, S., & Guibas, L. (2025). Img2CAD: Reverse Engineering 3D CAD Models from Images through VLM-Assisted Conditional Factorization. SIGGRAPH Asia 2025.