CVPR2021
Surrogate Gradient Field for Latent Space Manipulation
Minjun Li, Yanghua Jin, Huachun Zhu
摘要
Figure 1 . Our Surrogate Gradient Field (SGF) can edit images with diverse modalities of control by manipulating the latent codes of GANs. In the first row, we adjust the facial attributes of a person's photo. In the second row, we use natural language sentences to alter the color and the shape of a generated flower. In the last row, we edit keypoints of a generated anime character to modify head poses. We use StyleGAN2 [20] to generate the images above.