Deep learning to understand technical drawings

Some technical drawings are only available as raster images or in physical form. This is often but not only due to legacy reasons, like old floor plans. They traditionally need to be processed manually and are difficult to integrate into automated workflows, leading to significant extra work.

Former methods to analyze these raster images typically relied on extensive assumptions or had limited precision. With the field of deep learning and its contributions to image analysis, recent advancements have been made in tasks like vectorization, text detection, reconstruction of geometric shapes and understanding the role of the depicted objects.

Participants of this seminar learn about state-of-the-art methods based on neural networks that solve the various tasks associated with the analysis of technical drawings.

Registration

Please send an email to jasmin.hoffmann@kit.edu with your name and student ID (Matrikelnummer).

 

VortragsspracheDeutsch/Englisch