AttrLostGAN: Attribute Controlled Image Synthesis from Reconfigurable Layout and Style

Conditional image synthesis from layout has recently attracted much interest.\nPrevious approaches condition the generator on object locations as well as\nclass labels but lack fine-grained control over the diverse appearance aspects\nof individual objects. Gaining control over the image generation process is\nfundamental to build practical applications with a user-friendly interface. In\nthis paper, we propose a method for attribute controlled image synthesis from\nlayout which allows to specify the appearance of individual objects without\naffecting the rest of the image. We extend a state-of-the-art approach for\nlayout-to-image generation to additionally condition individual objects on\nattributes. We create and experiment on a synthetic, as well as the challenging\nVisual Genome dataset. Our qualitative and quantitative results show that our\nmethod can successfully control the fine-grained details of individual objects\nwhen modelling complex scenes with multiple objects. Source code, dataset and\npre-trained models are publicly available\n(https://github.com/stanifrolov/AttrLostGAN).\n