Unsupervised Coherent Video Cartoonization with Perceptual Motion Consistency

In recent years, creative content generations like style transfer and neural photo editing have attracted more attention. Among these, cartoonization of real-world scenes has promising applications in entertainment industry. Different from image translations focusing on improving the effect generated images, video additional requirements temporal consistency. this paper, we propose a spatially-adaptive semantic alignment framework with perceptual motion consistency for coherent an unsupervised manner. The module is designed to restore deformation structure caused by spatial information lost encoder-decoder architecture. Furthermore, introduce spatio-temporal correlative map as style-independent, global-aware regularization Deriving similarity measurement high-level features cartoon frames, it captures global beyond raw pixel-value optical flow. Besides, disentangles relationship domain-specific properties, which helps regularize without hurting effects images. Qualitative quantitative experiments demonstrate our method able generate highly stylistic consistent videos.