Generation of Building Plans Using ML and AI

Designing residential plans requires expertise and time. Traditional methods are prone to errors and are time-consuming. As a result, deep learning-based methods gained popularity in recent years. This research paper proposes a novel approach to generating building and residential plans using VAE and GANs. The project started with a VAE with Dall-e, which imposed more constraints on the output image. However, this led to pixelated images that were difficult to interpret. To overcome this, GANs were used, which required fewer constraints to produce more explicit output images. In GANs, two networks are trained simultaneously, a generator network that produces new samples and a discriminator network that evaluates the quality of the generated samples. Through adversarial training, the generator network learns to produce images that resemble the actual samples. The proposed GAN approach can overpower traditional residential plan design methods, producing accurate and high-resolution plans that are easier to interpret. However, using GANs is challenging, with instability in the training process leading to mode collapse or gradient vanishing. In conclusion, this research demonstrates the potential of deep learning-based methods for generating building or residential plans. The proposed approach using VAE and GANs can reduce the effort and time required to design a residential project, making it a valuable tool for architects and urban planners. However, further research is needed to address the challenges associated with GANs and improve the model's performance.