Object-Centric Image Generation from Layouts

We begin with the hypothesis that a model must be able to understand individual objects and relationships between in order generate complex scenes multiple well. Our layout-to-image-generation method, which we call Object-Centric Generative Adversarial Network (or OC-GAN), relies on novel Scene-Graph Similarity Module (SGSM). The SGSM learns representations of spatial scene, lead our model's improved layout-fidelity. also propose changes conditioning mechanism generator enhance its object instance-awareness. Apart from improving image quality, contributions mitigate two failure modes previous approaches: (1) spurious being generated without corresponding bounding boxes layout, (2) overlapping layout leading merged images. Extensive quantitative evaluation ablation studies demonstrate impact contributions, outperforming state-of-the-art approaches both COCO-Stuff Visual Genome datasets. Finally, address an important limitation metrics used works by introducing SceneFID -- object-centric adaptation popular Fréchet Inception Distance metric, is better suited for multi-object