Coarse-to-fine

Facial structure editing of portrait images is challenging given the facial variety, lack ground-truth, necessity jointly adjusting color and shape, requirement no visual artifacts. In this paper, we investigate how to perform chin as a case study structures. We present novel method that can automatically remove double effect in images. Our core idea train fine classification boundary latent space This be used edit appearance by manipulating code input image while preserving original features. To achieve such separation boundary, employ carefully designed training stage based on codes paired synthetic with without chin. testing stage, our handle only refinement subtle misalignment before after editing. model enables alteration neck region keeping other regions unchanged, guarantees rationality consistency characteristics. best knowledge, presents first effort towards an effective application for chins. validate efficacy efficiency approach through extensive experiments user studies.