Kwanggyoon Edward Seo

Hello World! I'm a Research Scientist at Flawless AI developing the next generation facial animation. I received my Ph.D. from the Graduate School of Culture Technology (GCST) at KAIST, where I was advised by Prof. Junyong Noh.

My research lies at the intersection of deep learning, computer vision, and computer graphics. Specifically, I am interested in generative AI, with a focus on animating 3D humans as well as manipulating images and videos.

Email  /  CV  /  Google Scholar  /  Github  /  LinkedIn

A novel deep learning-based virtual oculoplastic surgery simulation system that aims to improve the accuracy and quality of simulations by considering the anatomical structure and characteristics of the eye.

A method that enables direct retargeting between two facial meshes with different shapes and mesh structures.

An end-to-end approach to animating a 3D face mesh with arbitrary shape and triangulation from a given speech audio.

Propose a latent-based landmark detection and latent manipulation module to edit the emotion of portrait video that faithfully follows the original lip-synchronization or lip-contact.

Use a generative prior for identity agnostic audio-driven talking-head generation with emotion manipulation while trained on a single identity audio-visual dataset.

A novel method that can generate realistic speech animations of a 3D face from audio using multiple adaptive windows.

Generate stylized 3D face model with a single example paired mesh by finetuning a pre-trained surface deformation network.

Splatting deep generated features of pre-trained StyleGAN for cinemagraph generation.

Train a sketch generator with generated deep features of pre-trained StyleGAN to generate high-quality sketch images with limited data.

Generating and completing of 3D human RGB texture from a single image using sampling and refinement process from visible region.

A method to edit portrait video using a pre-trained StyleGAN using the video adaptaion and expression dynamics optimization.

A novel unsupervised learning method by reformulating the retargeting of 3D facial blendshape-based animation in the image domain.

A method that generates a virtual camera layout for both human and stylzed characters of a 3D animation scene by following the cinematic intention of a reference video.

A feed-forward neural network that can learn a semantic change of input images in a latent space to create the morphing effect by distilling the information of pre-trained GAN.

The source code of this website is from Jon Barron.

(last update: April.27 2025)