일시: 2021.11.05. 11:00
장소:온라인 세미나
발표자:김종유 박사님
주제:3D face geometry and texture reconstruction and application
개요: In this talk, I will introduce the 3D face reconstruction and related-vision tasks. The understanding of the human face is one of the most important modules in a number of industrial vision platforms. Especially, once we get a 3D face geometry can provide helpful information which can be applied to face recognition, face reenactment, face image synthesis, and so on. Since the DNN-based feedforward network has accuracy limitations, a multi-modal optimization-based approach can help to enhance the performance, where we deploy external face parsing and face alignment. In addition, we can also obtain high-fidelity complete facial textures by employing a generative adversarial network, which can play an important role in novel view synthesis or avatar generation. Based on the 3D face reconstruction pipeline, we generated synthetic data for masked face recognition and yielded a competitive performance in the ICCV 2021 workshop.
일시: 2021.11.05. 11:00
장소:온라인 세미나
발표자:김종유 박사님
주제:3D face geometry and texture reconstruction and application
개요: In this talk, I will introduce the 3D face reconstruction and related-vision tasks. The understanding of the human face is one of the most important modules in a number of industrial vision platforms. Especially, once we get a 3D face geometry can provide helpful information which can be applied to face recognition, face reenactment, face image synthesis, and so on. Since the DNN-based feedforward network has accuracy limitations, a multi-modal optimization-based approach can help to enhance the performance, where we deploy external face parsing and face alignment. In addition, we can also obtain high-fidelity complete facial textures by employing a generative adversarial network, which can play an important role in novel view synthesis or avatar generation. Based on the 3D face reconstruction pipeline, we generated synthetic data for masked face recognition and yielded a competitive performance in the ICCV 2021 workshop.