ABSTRACT
The image-to-image (I2I) translation refers to the process of transforming one image into another form/style according to the user’s requirements. It is a part of computer vision applications where the model aims to learn the relation between an input image and output image and maps them while preserving the essential features or attributes of the original image. Image-to-image translation applies many computer vision-related tasks such as generating images, pose estimation, segmentation of images, and making images high resolution. Image translation can be done using many methods. A commonly adopted technique for image-to-image translation is generative modelling. Currently, generative adversarial networks (GAN) are used for this process. Generative networks exhibit better results than other machine learning algorithms since they are faster and learn the distributions more quickly. The results generated by GAN are more realistic and have better image quality. GANs have a wide range of applications ranging from image synthesis, enhancement, style transfer, segmentation, and data extraction. This chapter explores image-to-image translation techniques, particularly the domain of style transfer. A few examples of GAN include deep convolutional GAN, CycleGAN, ConditionalGAN, and StarGAN.
