GenCAD is an image-conditional generative model that produces parametric CAD command sequences (CAD programs) from image inputs, rather than just 3D meshes or point clouds. The architecture combines four components: an autoregressive transformer encoder for CAD sequence representation, contrastive learning to align CAD command and image latent spaces, a latent diffusion model conditioned on images, and a decoder that converts latents into parametric CAD commands. The result is a full, editable CAD program that can be fed into a geometry kernel to produce a 3D solid model. GenCAD also supports image-conditional CAD retrieval from a collection of ~7000 programs.
436 Impressions