A Deep Learning-Based Approach to Generating Comprehensive Building Façades for Low-Rise Housing

In recent years, as machine learning has been widely studied in the field of architecture, scholars have demonstrated that computers can be used to learn graphical features building façade generation. However, existing deep generation yet generate only a single façade, without comprehensive five façades including roof. Moreover, most literature utilized Pix2Pix algorithm for experiments, failing attempt replace original generator with different experiments. This study addresses above issues by collecting and filtering entries from international Solar Decathlon (SD competition) obtain data set. Subsequently, low-rise residential model based on neural network was constructed training testing. At same time, U-net replaced three generators, U-net++, HRNet AttU-net, test results were obtained. The evaluated both subjective objective aspects it found AttU-net generative showed best performance such façades. is acceptable if there need fast