Master Thesis: Evaluation of Inpainting Methods for Floor Plan Detection
German Aerospace Center (DLR)
Datum: vor 4 Wochen
Stadt: Sankt Augustin, Nordrhein-Westfalen
Vertragstyp: Ganztags
Your Mission:
A digital twin is a virtual representation of a physical entity, process, or system. It enhances the monitoring, simulation, and optimization of real assets of infrastructures, strengthening their resilience. However, the generation of digital twins is extremely expensive due to the high amount of manual modeling work involved. Therefore, the development of automated techniques for generating virtual representations or digital twins holds significant importance. Information contained within technical drawings and data sheets, such as floor plans, circuit diagrams, and manufacturer specifications, might leverage the automated generation of Digital Twins. Given the high diversity and complexity of technical drawings and data sheets, AI-based methods are promising approaches for the digitalization.
The contours of the building structure are often covered by additional information as text or symbols. Hence removing text or symbols leads to a loss of valuable information by cutting out parts of this contours. AI-based techniques such as inpainting via GANs or Diffusion Models are promising approaches to address this challenge and reconstruct important information.
Therefore, the objective of this master thesis will be the development and evaluation of AI-based methods for text and symbol removal on floorplans, preserving essential information via inpainting.
Your tasks
A digital twin is a virtual representation of a physical entity, process, or system. It enhances the monitoring, simulation, and optimization of real assets of infrastructures, strengthening their resilience. However, the generation of digital twins is extremely expensive due to the high amount of manual modeling work involved. Therefore, the development of automated techniques for generating virtual representations or digital twins holds significant importance. Information contained within technical drawings and data sheets, such as floor plans, circuit diagrams, and manufacturer specifications, might leverage the automated generation of Digital Twins. Given the high diversity and complexity of technical drawings and data sheets, AI-based methods are promising approaches for the digitalization.
The contours of the building structure are often covered by additional information as text or symbols. Hence removing text or symbols leads to a loss of valuable information by cutting out parts of this contours. AI-based techniques such as inpainting via GANs or Diffusion Models are promising approaches to address this challenge and reconstruct important information.
Therefore, the objective of this master thesis will be the development and evaluation of AI-based methods for text and symbol removal on floorplans, preserving essential information via inpainting.
Your tasks
- exploring current state-of-the-art methods for inpainting and their applicability in the context of floorplans
- collecting and/or augmenting training data
- implementing and training of artificial neural networks for inpainting
- evaluating how inpainting can improve the digitization.
- currently enrolled as a Master student in computer science, mathematics, optical engineering or similar major
- experience in deep learning and/or computer vision is beneficial
- basic Python skills (ideally with respect to deep learning applications, e.g., PyTorch or Tensorflow)
- ability to work independently, good communication and teamwork skills