A Novel Video Coding Framework with GAN-based Face Generation for Videoconferencing
Soma Nagahara, Takafumi Katayama, Tian Song, Takashi Shimamoto · 2022
In this work, a novel video coding framework dedicated to videoconferencing is proposed. The proposed framework can improve coding efficiency by extracting facial landmarks and reconstructing the frame using the faces generated by GAN from the landmarks. By only transmitting the facial landmarks and key frames, the proposed coding structure achieves a very high coding efficiency, comparable to that of VVC down to QP52.