One-Shot Face Reenactment with 2D Facial Landmark Conditional Normalizing Flow

Abstract

Normalizing Flow (NF) has gained growing popularity in various image generation tasks. In this work, we develop a new method that enables the NF to control face generation, which has not been studied yet. To do so, we introduce several loss functions to facilitate stable training and inference while controlling face generation given a facial landmark. In our experiments, we evaluate the performance of the proposed method and show the capability of the NF in controlling the face generation task.