ZPdesu/Barbershop
GAN-based system for seamless hairstyle transfer and image blending using segmentation masks and latent-space manipulation.

Barbershop is a SIGGRAPH Asia 2021 publication implementing a GAN-inversion approach for image compositing. It enables transferring visual properties including hair, facial features, and details like moles from multiple reference images into a coherent output. The method operates in latent space to preserve detail and spatial information while conforming to a common segmentation mask, avoiding blending artifacts typical of pixel-space approaches.