SketchGraphs: A Large-Scale Dataset for Modeling Relational Geometry in Computer-Aided Design (2007.08506v1)

Abstract: Parametric computer-aided design (CAD) is the dominant paradigm in mechanical engineering for physical design. Distinguished by relational geometry, parametric CAD models begin as two-dimensional sketches consisting of geometric primitives (e.g., line segments, arcs) and explicit constraints between them (e.g., coincidence, perpendicularity) that form the basis for three-dimensional construction operations. Training machine learning models to reason about and synthesize parametric CAD designs has the potential to reduce design time and enable new design workflows. Additionally, parametric CAD designs can be viewed as instances of constraint programming and they offer a well-scoped test bed for exploring ideas in program synthesis and induction. To facilitate this research, we introduce SketchGraphs, a collection of 15 million sketches extracted from real-world CAD models coupled with an open-source data processing pipeline. Each sketch is represented as a geometric constraint graph where edges denote designer-imposed geometric relationships between primitives, the nodes of the graph. We demonstrate and establish benchmarks for two use cases of the dataset: generative modeling of sketches and conditional generation of likely constraints given unconstrained geometry.

• The paper introduces SketchGraphs, a large-scale dataset containing 15 million parametric CAD sketches represented as geometric constraint graphs to enable machine learning research in design automation.
• The SketchGraphs dataset enables benchmarks for generative modeling and constraint inference tasks, showing promise in predicting design sequences and geometric constraints.
• SketchGraphs supports applications beyond constraint inference, including enhancing image-to-CAD conversion and developing semantic representations for design data.

Overview of SketchGraphs: A Large-Scale Dataset for Modeling Relational Geometry in Computer-Aided Design

The paper "SketchGraphs: A Large-Scale Dataset for Modeling Relational Geometry in Computer-Aided Design" introduces an innovative dataset intended to advance the paper of parametric computer-aided design (CAD) through machine learning methodologies. The dataset, termed SketchGraphs, comprises 15 million 2D sketches extracted from CAD models on the Onshape platform, characterizing a rich amalgamation of geometric primitives and constraints.

Parametric CAD and Its Implications

Parametric CAD systems form the crux of modern mechanical design processes, employing geometric primitives – such as lines, circles, and arcs – that are interlinked through explicit constraints like perpendicularity, coincidence, or symmetry. These systems facilitate the creation of detailed 3D models from 2D sketches, where any alteration to a primitive's parameter propagates changes across dependent elements, thereby retaining the relational geometry.

The authors emphasize the potential of machine learning models to reshape CAD workflows by enhancing design efficiency and proposing novel design paradigms. Importantly, parametric CAD exemplifies instances of constraint programming, offering fertile ground for exploring machine learning applications in program synthesis and induction.

Introduction to SketchGraphs Dataset

The SketchGraphs dataset addresses a gap in existing CAD datasets by focusing on the geometric constraint structures that underpin parametric sketches rather than purely 3D volumetric shapes or images. Each sketch in the dataset is represented as a geometric constraint graph, where graph nodes are geometric primitives and edges denote constraints imposed during design.

The dataset's construction involved extracting these sketches from the Onshape CAD platform, ensuring each sketch incorporates at least one primitive and constraint, thereby maintaining its relevance to modelling relational geometry. The resultant dataset is proposed to benefit target applications such as conditional completion of sketches and inference of constraints from shapes, expanding traditional processes predominantly reliant on heuristic approaches.

Computational and Theoretical Exploration

Initial results establish benchmarks in generative modeling and constraint inference using SketchGraphs. In generative modeling, models predict construction sequences which can assist users by suggesting subsequent design steps or assessing design plausibility. For constraint inference, they aim to automate the process of determining 'natural' constraints for a given set of primitives, a task akin to relational graph modeling and link prediction.

The authors describe an approach leveraging message-passing networks for constraint prediction, achieving an average precision and recall of 0.74, indicating strong performance in reconstructing designer-imposed relationships from primitive data alone.

Prospect and Future Applications

The utility of SketchGraphs extends to various domains, including enhancing CAD systems through image-based inference. By portraying sketches not as static images but as structured data tied to relational constraints, machine learning models can facilitate the transition from image descriptions to CAD models, enabling further advancements in mechanical design and human-computer interaction.

Moreover, SketchGraphs potentially aids in the development of semantic representations for CAD data, providing analogy to latent semantic representations in NLP, which could power innovative search, categorization, and recommendation systems for design patterns.

Conclusion

Ultimately, the SketchGraphs dataset is a strategic resource purposed to guide researchers in deploying machine learning tactics within the CAD domain, addressing challenges in structural reasoning and promoting advancements in program synthesis. With CAD at the intersection of graphical and computational disciplines, SketchGraphs marks a significant step towards redefining design methodologies via artificial intelligence. While the paper outlines an extensive scope, ongoing research is poised to expand upon this foundation, exploring comprehensive 3D design processes and extending benchmarks for CAD-based AI applications.