Method and apparatus for generating face rotation image

A method for training a generative adversarial network, applied to a computer device, wherein the generative adversarial network comprises a face generator network and a plurality of coupled adversarial discriminator networks, the coupled adversarial discriminator networks comprise at least a first discriminator network and a second discriminator network, and the method comprises:

• receiving a face image and a face rotation image, wherein the face image and the face rotation image are images before and after a same face is rotated;
• performing pose encoding on the face image based on two or more landmarks in the face image, to obtain pose encoded images of the face image;
• performing pose encoding on the face rotation image based on two or more landmarks in the face rotation image, to obtain pose encoded images of the face rotation image;
• inputting the face image, the pose encoded images of the face image, and the pose encoded images of the face rotation image into the face generator network, to generate a predicted face rotation image;
• inputting the face image, the face rotation image, and the predicted face rotation image into the first discriminator network, to obtain a first loss;
• inputting the face rotation image, the pose encoded images of the face rotation image, and the predicted face rotation image into the second discriminator network, to obtain a second loss;
• updating the face generator network, the first discriminator network, and the second discriminator network based on a total loss of the generative adversarial network, wherein the total loss of the generative adversarial network is obtained based on a weighted sum of the first loss and the second loss;
• and
• outputting the trained face generator network after the total loss of the generative adversarial network is converged.

A device for generating a face rotation image, comprising a processor and a memory, wherein the memory is configured to store a program instruction, and the processor is configured to execute the program instruction to perform a method for generating a face rotation image, the method comprising:

• receiving a face image;
• performing pose encoding on the face image based on two or more landmarks in the face image, to obtain pose encoded images of the face image;
• obtaining a plurality of first training images from a training data set based on a face rotation angle, wherein each of the plurality of first training images comprises a face, and a presented rotation angle of the face comprised in each of the plurality of first training images is the face rotation angle;
• performing pose encoding on a target face image based on two or more landmarks in the target face image, to obtain pose encoded images of the target face image, wherein the target face image is obtained based on the plurality of first training images;
• generating a to-be-input signal based on the face image, the pose encoded images of the face image, and the pose encoded images of the target face image, wherein a size of the face image, a size of the pose encoded images of the face image, and a size of the pose encoded images of the target face image are the same;
• and
• inputting the to-be-input signal into a face rotation image generative model, to obtain a face rotation image.