A Comparative Analysis of Satellite Image to Map Image Translation Using GANs

Image-to-image translation is a challenging task that aims to convert an input image from one domain to another while preserving its semantic content. In this comparative study, we examine four state-of-the-art Generative Adversarial Network (GAN) architectures: Pix2Pix, CycleGAN, and DualGAN, for the specific task of translating satellite images to map types and examine how they are used in the classification of land cover, urban planning, disaster management, and environmental monitoring. Using several measures such as Inception score, Fréchet Inception Distance, Pixel Accuracy, and Structural Similarity, we assess how models are producing maps and compare their respective performances. This thorough comparative study aims to provide insights into the strengths and weaknesses of Pix2Pix, CycleGAN, and DualGAN in the context of satellite image-to-map translation. Our findings can aid researchers and practitioners in selecting the most appropriate GAN architecture for this task, ultimately advancing the field of image-to-image translation for satellite imagery applications.