A CAD Drawing Restoration Method Based on Swin-Transformer and MAE

With the wide application of CAD drawings in architecture, engineering and manufacturing, the drawing restoration task has become an important challenge to improve the quality and accuracy of drawings. In this paper, we propose an improved algorithm based on Swin-Transformer and Masked Auto-Encoder (Swin-MAE), aiming to solve the problems of structural consistency and detail restoration in CAD drawing restoration. Swin-MAE combines the local and global information modelling capabilities of Swin- Transformer, and efficiently captures long-distance dependencies and complex structures in drawings by using the window self-attention mechanism and window panning strategy to effectively capture long-distance dependencies and complex structures in drawings, and at the same time restores missing regions using the occlusion learning mechanism of MAE. Experiments show that Swin-MAE has significant advantages over traditional methods in the CAD drawing repair task, and Swin-MAE can also achieve high repair accuracy and training efficiency when using a single block decoder. The method provides a new efficient solution for CAD drawing repair and has a wide range of application prospects.