This paper presents a novel method for generating high-resolution, photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). The authors address two main challenges: generating high-resolution images with GANs and improving the details and textures in high-resolution results. They achieve this by introducing a coarse-to-fine generator, multi-scale discriminators, and an improved adversarial loss function. Additionally, they incorporate instance-level object segmentation information to enable flexible object manipulations, such as adding or removing objects and changing their categories. They also propose a method to generate diverse results given the same input, allowing users to interactively edit the appearance of objects. The method is evaluated through quantitative comparisons and human perceptual studies, demonstrating superior performance in terms of image quality and realism compared to existing methods. The authors also provide interactive demos and discuss the potential applications of their method in various domains.This paper presents a novel method for generating high-resolution, photo-realistic images from semantic label maps using conditional generative adversarial networks (conditional GANs). The authors address two main challenges: generating high-resolution images with GANs and improving the details and textures in high-resolution results. They achieve this by introducing a coarse-to-fine generator, multi-scale discriminators, and an improved adversarial loss function. Additionally, they incorporate instance-level object segmentation information to enable flexible object manipulations, such as adding or removing objects and changing their categories. They also propose a method to generate diverse results given the same input, allowing users to interactively edit the appearance of objects. The method is evaluated through quantitative comparisons and human perceptual studies, demonstrating superior performance in terms of image quality and realism compared to existing methods. The authors also provide interactive demos and discuss the potential applications of their method in various domains.