Dr. Junxuan Bai

Avatars in NeRF

LiteNeRFAvatar (Will Be Available Soon)
LiteNeRFAvatar: A lightweight NeRF with local feature learning for dynamic human avatar (Pattern Recognition, 2026).

Human Pose Estimation

DGFormer
DGFormer: Dynamic Graph Transformer for 3D Human Pose Estimation (Pattern Recognition, 2024).

Motion Prediction

KD-Former
KD-Former: Kinematic and dynamic coupled transformer network for 3D human motion prediction (Pattern Recognition, 2023).

Motion Synthesis

Dance-Synthesis-Project
In the related paper, we propose a keyframe-based motion generation network based on multiple constraints, which can achieve diverse dance synthesis via learned knowledge. The work is published at Computer Graphics Forum (Special Issue--Pacific Graphics 2021).