Architectural Map Generation using GANs

In response to the client's need for diverse map image generation from complex architectural diagrams, we developed a solution using advanced machine learning techniques. The challenge lay in refining the dataset to remove unwanted objects and augmenting it to facilitate the creation of varied map images. By implementing a tailored solution, we aimed to meet the client's requirements efficiently while ensuring scalability and customization.


The client provided architectural diagrams containing diverse objects, complicating the process of generating map images. They sought to enhance the diversity of their dataset to enable the AI model to produce a broader range of map images. Additionally, the client required customization options to tailor the generated images to their specific needs. Addressing these challenges while maintaining efficiency and scalability was essential to meeting the client’s expectations.


To meet the client’s needs, we used a combination of Generative Adversarial Networks (GANs) and image-to-image translation models. Firstly, we carefully cleaned up the architectural diagrams to get rid of any unwanted objects, making sure our dataset was top-notch for training. Then, we beefed up the dataset to make it more diverse, allowing the AI model to create a wider variety of map images. For training, we made the most of Google Colab, which helped us streamline the process, saving time and resources. We also tweaked the model settings to make sure it performed at its best, delivering accurate results. Our solution allowed for customized outputs, perfectly tailored to the client’s specific preferences and requirements.

Features & Benefits

Enhanced Data Diversity

By cleaning the architectural diagrams and generating new data, our solution significantly increased the diversity of the dataset, enabling the AI model to produce a wide range of map images.

Customizable Output

The image-to-image translation model provided flexibility in adjusting the output according to the client’s specific requirements, ensuring tailored results.

Efficient Training

Leveraging Google Colab for training streamlined the process, optimizing resource utilization and reducing development time.


The developed model could generate a large number of images, providing scalability to accommodate future expansion or changes in client demands.

Technologies Used to Build the Solution

  • Python: Used for programming the solution and implementing machine learning algorithms.
  • Generative Adversarial Networks (GANs): Employed for generating realistic map images by training a generator model against a discriminator model.
  • Image-to-Image Translation Model: Utilized to clean the architectural diagrams by removing unwanted objects and generating new data based on the cleaned images.
  • Google Colab: Chosen as the platform for training the model, offering scalable computing resources and collaborative features for efficient development.