SketchDNN: Joint Continuous-Discrete Diffusion for CAD Sketch Generation

TL;DR

SketchDNN introduces a novel diffusion-based generative model for CAD sketches that effectively handles continuous parameters and discrete labels, significantly improving generation quality and setting new state-of-the-art results.

Contribution

The paper proposes Gaussian-Softmax diffusion for joint modeling of continuous and discrete variables in CAD sketch generation, addressing heterogeneity and permutation invariance challenges.

Findings

Reduced FID from 16.04 to 7.80

Lowered NLL from 84.8 to 81.33

Achieved new state-of-the-art in CAD sketch generation

Abstract

We present SketchDNN, a generative model for synthesizing CAD sketches that jointly models both continuous parameters and discrete class labels through a unified continuous-discrete diffusion process. Our core innovation is Gaussian-Softmax diffusion, where logits perturbed with Gaussian noise are projected onto the probability simplex via a softmax transformation, facilitating blended class labels for discrete variables. This formulation addresses 2 key challenges, namely, the heterogeneity of primitive parameterizations and the permutation invariance of primitives in CAD sketches. Our approach significantly improves generation quality, reducing Fr\'echet Inception Distance (FID) from 16.04 to 7.80 and negative log-likelihood (NLL) from 84.8 to 81.33, establishing a new state-of-the-art in CAD sketch generation on the SketchGraphs dataset.