Sketch2CAD: Generative Adversarial Network for Automated Conversion of Hand-Drawn Sketches to Parametric CAD Models
Xiaogang Wang, Liu YunCong, Yu Zhang
AI summary
Problem
Converting imprecise hand-drawn sketches into precise, parametric CAD models is labor-intensive and error-prone, while existing automated methods struggle with poor sketch quality, variable geometric primitives, and implicit constraint extraction.
Approach
A modified CycleGAN framework simultaneously corrects hand-drawn sketches into clean CAD-like images and performs pixel-level semantic segmentation to identify geometric primitives, followed by a post-processing pipeline that extracts parametric data and infers geometric constraints.
Key results
- Achieves 94.56% primitive accuracy and 81.35% constraint recognition recall
- Outperforms Vitruvion and PPI-Net across all evaluation metrics
- Successfully extracts parametric primitives and infers geometric constraints for direct CAD integration
- Demonstrates robust generalization through targeted geometric data augmentation
Why it matters
Drastically reduces manual drafting effort while maintaining engineering precision, enabling faster design iteration for robotics and industrial prototyping.
Abstract
This paper addresses the labor-intensive process of converting imprecise hand-drawn sketches into precise, parametric CAD sketches. We present Sketch2CAD, a novel deep learning framework that leverages generative adversarial networks (GANs) to automate this conversion. Our approach consists of two main stages: first, a sketch correction module transforms freehand sketches into clean, standardized CAD-like sketches; second, a semantic segmentation module parses the generated sketches to identify and classify geometric primitives (lines, circles, arcs, points). We further introduce an optimized post-processing algorithm that extracts parametric primitives and infers geometric constraints from the segmentation results, enabling direct integration with commercial CAD software. Ex- tensive experiments demonstrate that our method significantly outperforms state-of-the-art approaches in both primitive accu- racy (94.56%) and constraint recognition. This work provides a robust solution that reduces manual effort in CAD drafting while maintaining engineering precision, particularly suitable for robotics applications requiring rapid prototyping and design iteration.