GameCAD Dataset: A Multi-View Image-to-CAD Dataset for Editable Game Asset Modeling

The GameCAD Dataset is a game-oriented CAD dataset designed for editable 3D asset modeling from multi-view images. It focuses on hard-surface game asset models and provides paired visual and programmatic CAD representations. Each sample contains rendered multi-view images and corresponding CADQuery code, enabling models to be reconstructed as editable parametric CAD programs rather than static triangle meshes. The dataset covers representative game asset categories and includes geometric structures commonly used in game production, such as weapons, tools, props, equipment components, and stylized mechanical parts. By linking visual observations with executable CAD programs, the dataset supports research on image-to-CAD generation, CAD reverse engineering, and code-based 3D modeling. It can be used to train and evaluate models that infer structured modeling operations and continuous geometric parameters from visual inputs. This dataset provides a foundation for scalable, editable, and production-oriented game asset creation.