TriView-CAD: A Three-View Mechanical Engineering Drawing Dataset for Consistency Verification

TriView-CAD is a machine-vision dataset developed for consistency verification of three-view mechanical engineering drawings. The dataset contains orthographic drawing images of mechanical parts, including front, top, and side views extracted from CAD-generated engineering drawings. It is intended to support research on engineering drawing understanding, cross-view representation learning, and automatic consistency inspection. The original drawings were preprocessed to separate the three orthographic views and remove irrelevant borders, residual lines, and background components. Each sample is organized as a three-view combination with a corresponding consistency label. Consistent samples contain three views belonging to the same mechanical part, whereas inconsistent samples are constructed by combining views from different parts. To prevent data leakage, the dataset is divided into training, validation, and test sets according to part identifiers before inconsistent samples are generated, ensuring that the same part does not appear across different subsets. The repository includes the processed engineering drawing images, sample labels, fixed training/validation/test splits, and documentation describing the file organization and naming conventions. The dataset can be used for binary consistency classification, multi-view feature learning, engineering drawing analysis, and related computer-vision tasks.