Cycle consistent generative adversarial network for unpaired image translation

Abstract

Unpaired image translation has become an important research topic in computer vision due to the difficulty of obtaining paired datasets. Cycle-consistent generative adversarial networks provide an effective solution by learning mappings between two image domains without paired supervision. This paper analyzes the CycleGAN model, including its architecture and training strategy. Experiments are conducted on benchmark datasets such as horse-to-zebra and seasonal translation tasks, evaluated using SSIM and PSNR. The results show that CycleGAN generates visually realistic images while preserving structural content and reduces reliance on labeled data compared with supervised methods.