AFF‐UNIT: Adaptive feature fusion for unsupervised image‐to‐image translation

The task of image-to-image translation is to generate images closer the target domain style while preserving significant features original image. This paper contends an adaptive feature fusion method for unsupervised image translation. proposed architecture, termed as AFF-UNIT, based on a compact network structure further improve quality generated images. First all, extraction module proposed, which combines low-level fine-grained information and high-level semantic obtain maps with richer information. At same time, feature-similarity loss guide extract that are more conducive improving result. In addition, AFF-UNIT reuses in generator discriminator simplify framework. Extensive experiments five popular benchmarks demonstrate superior performance over state-of-the-art methods terms FID, KID, IS, also human preference. Comprehensive ablation studies carried out isolate validity each component.