CDGAN: Cyclic Discriminative Generative Adversarial Networks for Image-to-Image Transformation

Generative Adversarial Networks (GANs) have facilitated a new direction to\ntackle the image-to-image transformation problem. Different GANs use generator\nand discriminator networks with different losses in the objective function.\nStill there is a gap to fill in terms of both the quality of the generated\nimages and close to the ground truth images. In this work, we introduce a new\nImage-to-Image Transformation network named Cyclic Discriminative Generative\nAdversarial Networks (CDGAN) that fills the above mentioned gaps. The proposed\nCDGAN generates high quality and more realistic images by incorporating the\nadditional discriminator networks for cycled images in addition to the original\narchitecture of the CycleGAN. The proposed CDGAN is tested over three\nimage-to-image transformation datasets. The quantitative and qualitative\nresults are analyzed and compared with the state-of-the-art methods. The\nproposed CDGAN method outperforms the state-of-the-art methods when compared\nover the three baseline Image-to-Image transformation datasets. The code is\navailable at https://github.com/KishanKancharagunta/CDGAN.\n