This thesis adopts an interdisciplinary research that integrates architectural design and artificial intelligence techniques to assist architects in performing early conceptual design. We consider the use of deep learning models known as generative adversarial networks (GANs) to translate the sketches hand-drawn by architects into rendered architectural images to provide architects with multiple design options or inspirations. These rendered architectural images and imitative sketches are used to build deep learning datasets. By training the models based on pix2pix and CycleGAN with the prepared datasets, novel sketches drawn by the architects can be promptly rendered and generated to facilitate iterative ideas in the early design stage. The results of different models are evaluated con...[ Read more ]

This thesis adopts an interdisciplinary research that integrates architectural design and artificial intelligence techniques to assist architects in performing early conceptual design. We consider the use of deep learning models known as generative adversarial networks (GANs) to translate the sketches hand-drawn by architects into rendered architectural images to provide architects with multiple design options or inspirations. These rendered architectural images and imitative sketches are used to build deep learning datasets. By training the models based on pix2pix and CycleGAN with the prepared datasets, novel sketches drawn by the architects can be promptly rendered and generated to facilitate iterative ideas in the early design stage. The results of different models are evaluated considering the structural similarity index and Frechet inception distance metrics. The proposed novel approach can effectively reflect architects’ early creative inspiration through simple sketches within seconds. Without the need to repeatedly use design software for complex modeling and renderings, this approach can stimulate architects with additional creative possibilities suggested promptly. This work provides novel insights regarding the combination of artificial intelligence rationality and architects’ creative thinking to facilitate architectural design.