Seokhyeon Hong

Hello, world! I am a Ph.D. student at KAIST Visual MediaLab, advised by Prof. Junyong Noh. My research interests lie in computer graphics and character animation, including generation, editing, in-betweening, retargeting, rigging, and all the fun stuff that makes characters move.

Email  /  CV  /  Scholar  /  Github  /  Blog  /  YouTube

Publications

Skeleton-aware latent diffusion that incorporates interactions between skeletal joints, motion frames, and textual words for text-driven motion generation and zero-shot editing.

Motion in-betweening for arbitrary characters using video diffusion models fine-tuned with ICAdapt and motion-video mimicking, recuding dependency on character-specific datasets.

Rigging and skinning 3D character meshes by leveraging cross-attention modules and 2D generative priors for robust generalization across diverse skeletal and mesh configurations.

Retargeting interaction motions of two skinned characters via sparse mesh-agnostic geometry representations and spatio-cooperative transformers.

Keyframe prediction using two-stage hierarchical transformers for long-term motion in-betweening.

Training a recurrent motion refiner for neural locomotion stitching.

Projects

A python library for motion data processing with nice-looking visualization using OpenGL. It provides several functionalities not only motion-related functions (e.g. forward kinematics) but also auxiliary functions such as rotation conversion, file I/O, visualization with skinned mesh, and more. Most of my research projects have utilized this library for data processing and visualization.