GOS: A LARGE-SCALE ANNOTATED OUTDOOR SCENE SYNTHETIC DATASET

Scene editing has attracted increasing research interests owing to its valuable applications in the field of photography and entertainment. With style-based GAN being proposed, images can be reasonably edited on specific semantics by manipulating in latent space of the generator. However, existing datasets cannot satisfy the demands of large amounts of diverse data and rich semantic annotations at the same time, which makes the existing method difficult to edit on the content of outdoor scene images. To address these problems, we propose a large-scale, diverse synthetic dataset called "GOS dataset" generated based on a video game, which contains fine-grained semantic annotations. Extensive experiments show that utilizing the features obtained from the annotations of our dataset achieves better performance in outdoor scene editing, especially for distance and viewpoint of scenes, which indicates the extracted features have a certain generalization capability.