StarGAN: Unified Generative Adversarial Networks for Multi-domain Image-to-Image Translation

Recent studies achieved remarkable success in image-to-image translation for two domains, but existing approaches have limited scalability and robustness when handling more than two domains because a different model must be built independently for every pair of image domains. To address this, the authors propose StarGAN, a novel and scalable approach that can perform image-to-image translations for multiple domains using only a single model, whose unified architecture allows simultaneous training of multiple datasets with different domains within one network.

This unified design leads to superior quality of translated images compared to existing models, along with the novel capability of flexibly translating an input image to any desired target domain. The authors empirically demonstrate the effectiveness of the approach on a facial attribute transfer task and a facial expression synthesis task, establishing StarGAN as a unified solution for multi-domain translation.