GenCAD
7 hours ago
- #CAD generation
- #machine learning
- #parametric modeling
- GenCAD is an image-conditional generative model that produces parametric CAD command sequences (CAD programs), which can be converted into 3D solid models using a geometry kernel.
- The model addresses the complexity of CAD data structures like B-rep by generating entire CAD command histories, ensuring accuracy and modifiability for engineering tasks, unlike simpler representations such as meshes or point clouds.
- The architecture combines an autoregressive transformer encoder for latent CAD sequence representation, contrastive learning for joint CAD-image representations, a latent diffusion model for conditioned generation, and a decoder to output parametric CAD commands.
- GenCAD enables creating CAD models from image renderings, generating multiple CAD samples per input image, and offers image-conditional CAD retrieval, as demonstrated by retrieving top-3 CAD programs from a collection of about 7,000.