Design and Implementation of Real-time Face Changing based on wrapGAN Model
Jiawang Chen · 2021 IEEE 3rd Eurasia Conference on IOT, Communication and Engineering (ECICE) · 2021
This research aims to design and implement real-time generation of the target face in the monocular camera to achieve face change using Image-to-Image Translation with Conditional Adversarial Nets. The goal is to perform deep learning based on two sets of image domains and obtain the mapping relationship between input images and output images. However, for many face-changing tasks with specific identities, matching training data cannot be provided. The result realizes the acquisition and production of specific character data sets. The annotation process is reduced by using the modified wrapGAN network model for training and finally pruning the GAN model to leave the generative model. Then, real-time tracking of the face through the monocular camera is realized and the target face according to the generative model is generated. The model in this article retains user identity information, not relying on traditional 3dmm. It uses the wrapGAN model as a general solution for face changing. The data set comes from the network, of which 80% is used for training and 20% is used for testing. Compared with the previous methods, the superiority of our method is proved.