Search papers, labs, and topics across Lattice.
The paper introduces CADEvolve, an evolution-based pipeline that leverages VLMs to generate complex CAD programs from simple primitives through iterative edits and validations. This addresses the data scarcity of complex CAD programs with multi-operation composition and design intent, which are lacking in existing datasets. The resulting CADEvolve dataset, containing 1.3 million scripts paired with rendered geometry, enables fine-tuning of VLMs that achieve state-of-the-art performance on Image2CAD tasks.
Forget sketch-extrude sequences: CADEvolve uses program evolution to create a dataset of 1.3M complex, executable CAD programs, enabling state-of-the-art Image2CAD performance when used to fine-tune VLMs.
Computer-Aided Design (CAD) delivers rapid, editable modeling for engineering and manufacturing. Recent AI progress now makes full automation feasible for various CAD tasks. However, progress is bottlenecked by data: public corpora mostly contain sketch-extrude sequences, lack complex operations, multi-operation composition and design intent, and thus hinder effective fine-tuning. Attempts to bypass this with frozen VLMs often yield simple or invalid programs due to limited 3D grounding in current foundation models. We present CADEvolve, an evolution-based pipeline and dataset that starts from simple primitives and, via VLM-guided edits and validations, incrementally grows CAD programs toward industrial-grade complexity. The result is 8k complex parts expressed as executable CadQuery parametric generators. After multi-stage post-processing and augmentation, we obtain a unified dataset of 1.3m scripts paired with rendered geometry and exercising the full CadQuery operation set. A VLM fine-tuned on CADEvolve achieves state-of-the-art results on the Image2CAD task across the DeepCAD, Fusion 360, and MCB benchmarks.
