FloorplanGAN: Vector residential floorplan adversarial generation

An architectural floorplan is a class of drawings that reflects the layout of rooms. The difference between a floorplan and a natural image and its dual features as both a vector graphic and a raster image makes it difficult to be generated by conventional deep neural generative models. We propose an adversarial generative framework that combines vector generation and raster discrimination for residential floorplan generation tasks. The floorplan is first generated in vector format with room areas as constraints and then discriminated in raster format visually using convolutional layers. A Differentiable Renderer connects the gap between the Vector Generator and Raster Discriminator. A self-attention mechanism is utilized to capture the interrelations of rooms in each floorplan. Experiments were conducted to demonstrate the feasibility of the proposed FloorplanGAN. In addition, we evaluated the effectiveness of generation based on diverse objective metrics and a user study. The code is available here: https://github.com/luozn15/FloorplanGAN.