Unsupervised many‐to‐many image‐to‐image translation across multiple domains

Unsupervised multi-domain image-to-image translation aims to synthesize images among multiple domains without labelled data, which is more general and complicated than one-to-one image mapping. However, existing methods mainly focus on reducing the large costs of modelling do not pay enough attention quality generated images. In some target domains, their results may be expected or even cause model collapse. To improve quality, an effective many-to-many mapping framework for unsupervised proposed. There are two key aspects proposed method. The first a architecture with only one domain-shared encoder several domain-specialized decoders effectively simultaneously translate across domains. second constraints extended from mappings further help generation. All evaluations demonstrate that superior provides solution translation.