Cross-domain Image Generation: Style Transfer and Image-to-image Translation

Abstract

. With the development of deep learning, cross-domain image generation has become an important research direction in computer vision, with broad application prospects in artistic creation and image editing. This paper summarizes the research results, principles, and applications of cross-domain image generation from two aspects: style transfer and image-to-image translation. Through comparative analysis of core technologies such as conditional generative adversarial networks (GANs), recurrent consistency GANs, and attention mechanisms, this paper outlines the development process of cross-domain image generation technology. From this can conclude that style transfer focuses on the art and image translation on the semantic maps. These two algorithms differ in loss functions, networks, and training but both suffer from generation quality, robustness, and computational cost. Through comparative analysis, this paper finds that style transfer emphasizes artistic rendering through content–style decoupling, whereas image-to-image translation focuses on semantic consistency across domains using paired or unpaired data. Although significant progress has been achieved in generation quality, diversity, and controllability, challenges remain in computational efficiency, training stability, generalization ability, and evaluation metrics. Recent advances in diffusion models further demonstrate improved generation fidelity and stability, suggesting a shift from adversarial learning to probabilistic generative modeling. Overall, this work provides a comprehensive overview of current techniques and identifies future research directions, including lightweight model design, multimodal integration, and more robust and interpretable evaluation frameworks.