Abstract
Generative adversarial networks (GANs) have made remarkable success in image generations. However, how to deal with the multi-domain particular scenes generation, which converts specific object images to different reasonable scene domians, is still an open problem. In this paper, we propose a multi-domain particular scene generation model named PSceneGAN (Particular Scene Generative Adversarial Nets) that is a novel dual-condition GAN. PSceneGAN is the first model to achieve one-to-many specific scene generation under the guidance of semantics using only one model. In addititon, we collect and label a novel high-quality clothing data set named DRESS and use it to verify our PSceneGAN through a challenging task. The results show that PSceneGAN not only accurately generates corresponding reasonable scene images according to input scene and semantic descriptions, but also achieves desired results in quantitative and qualitative evaluation, among which frechet inception distance (FID) and inception score (IS) are 25.40 and 36.24, respectively.
Highlights
The task of generating a particular scene is to convert an object image to a specific reasonable scene domian
Our work is similar to StarGAN, but based on it, the difference is that StarGAN generates directly on the original input image and the labels are for the input images, while our labels are for the target scene images
The other method we used is to randomly select 200 input scenes from the test set firstly, StarGAN and PSceneGAN generate the target scene images described by the labels respectively (Since pix2pix and CycleGAN have not added semantics, they do not participate in the third item)
Summary
The task of generating a particular scene is to convert an object image to a specific reasonable scene domian. It still has no ability to generate multiple reasonable scene domains for a given input image under semantic control. Many recent related works have shown that corresponding images can be generated using semantic control [12], [13], while they generate matching images directly from the text description. It is difficult for the existing methods to solve the generation of particular scenes. We propose a novel two-condition generative adversarial network PSceneGAN to generate a specific scenario It uses semantic tags and target scenes as conditions to generate multiple scenes for an input image, which effectively implements semantic control for the image generation. We present a data set DRESS containing a large number of corresponding scene images and tags, which includes dresses, models and attribute labels
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
