Jingzhong Lin (林靖众)

Unified human–camera generation with explicit control over camera translation intensity.

Flow matching and physics-guided reinforcement learning for plausible human motion capture.

ReactDance: Hierarchical Representation for High-Fidelity and Coherent Long-Form Reactive Dance Generation

ReactDance enables fine-grained duet interaction via hierarchical latent representations and ensures long-term coherence through temporal alignment.

A text-to-motion framework that integrates OCEAN personality traits and Laban-based guidance to generate psychologically consistent and personality-distinct human motions.

@article{LiLZQWH25,
  author = {Xinru Li and Jingzhong Lin and Bohao Zhang and Yuanyuan Qi and Changbo Wang and Gaoqi He},
  title = {EmoDiffGes: Emotion-Aware Co-Speech Gesture Generation with Progressive Synergistic Diffusion},
  year = {2025},
  journal = {Computer Graphics Forum (PG 2025)}
}

EmoDiffGes, a diffusion-based framework unifying dynamic emotion conditioning and part-aware synergistic modeling.

@inproceedings{10.1145/3746027.3755855,
  author = {Bohao Zhang, Haoxin Xu, Jingzhong Lin, Changbo Wang, Gaoqi He},
  title = {Regulatory Focus Theory Induced Micro-Expression Analysis with Structured Representation Learning},
  year = {2025},
  booktitle = {ACM MM}
}

Ac4AU, an apex-free network for action unit detection that addresses data leakage concerns raised in CD6ME and eliminates the need for manually annotated apex frames.

BITE is an open-source local-first workflow for structured paper analysis, research memory, and knowledge-grounded research agents.