Yi Liu's Homepage. 😀

Jiading Campus, Tongji University

4800 Cao'an Road, Jiading District

Shanghai 201804, China

Yi Liu (Liu Yi, 刘毅) is a master’s student in Computer Science and Technology at the ADMIS Lab, Tongji University, advised by Associate Professor Wengen Li. He received his bachelor’s degree from Tongji University in 2024 and expects to complete his master’s degree in March 2027.

His research focuses on efficient and controllable generative models, particularly diffusion models and flow matching, for image restoration, image and video generation, and digital-human motion generation.

His main interests are:

Efficient generation: few-step sampling, lightweight architectures, and hardware-aware acceleration.
Conditional control: restoration guidance and a controllable balance between faithful reconstruction and realistic generation.
Long-sequence modeling: history-conditioned generation with temporal coherence and consistent motion style.

He is currently a research intern at ByteDance (since May 2026). Previously, he worked at Alibaba Cloud’s Tongyi Lab on long-horizon co-speech motion generation (June 2025–May 2026) and visited SCI Lab, Westlake University, mentored by Professor Xin Yuan (February–June 2024).

His long-term goal is to make controllable generation fast and reliable enough for real-time, interactive applications through advances in both algorithms and hardware-aware implementation. He received the National Scholarship for Graduate Students in 2025.

📧 Contact: liuyi61@tongji.edu.cn or liuyi2052697@foxmail.com.

news

Apr 16, 2025	Yi Liu’s website has been successfully launched and is now fully operational 🥳🤠. The site is accessible online, and all core features have been implemented. Visitors can explore the available content and functionalities as intended.

latest posts

Jan 20, 2025	离散型流匹配模型（3）
Jan 19, 2025	离散型流匹配模型（2）
Jan 05, 2025	离散型流匹配模型（1）

selected publications

NeurIPS

Past as State, Present as Attention: Persistent-State Blockwise Flow Matching for Long-Horizon Co-Speech Motion Generation

Yi Liu, Xiangyue Zhang, Jianfang Li, and 3 more authors

2026

Bib PDF Video Website

@misc{liu2026paspa,
  title = {Past as State, Present as Attention: Persistent-State Blockwise Flow Matching for Long-Horizon Co-Speech Motion Generation},
  author = {Liu, Yi and Zhang, Xiangyue and Li, Jianfang and Ma, Jia and He, Yisheng and Ren, Jianqiang},
  year = {2026},
}

arXiv

Disentangling Generation and Regression in Stochastic Interpolants for Controllable Image Restoration

Yi Liu, Jia Ma, Wengen Li, and 3 more authors

arXiv preprint arXiv:2605.21381, 2026

arXiv Bib PDF

@article{liu2026disi,
  title = {Disentangling Generation and Regression in Stochastic Interpolants for Controllable Image Restoration},
  author = {Liu, Yi and Ma, Jia and Li, Wengen and Guan, Jihong and Zhou, Shuigeng and Zhang, Yichao},
  journal = {arXiv preprint arXiv:2605.21381},
  year = {2026},
  url = {https://arxiv.org/abs/2605.21381}
}

arXiv

SADER: Structure-Aware Diffusion Framework with DEterministic Resampling for Multi-Temporal Remote Sensing Cloud Removal

Yifan Zhang, Qian Chen, Yi Liu, and 2 more authors

arXiv preprint arXiv:2602.00536, 2026

Yi Liu is a co-first author.

arXiv Bib PDF

@article{zhang2026sader,
  title = {SADER: Structure-Aware Diffusion Framework with DEterministic Resampling for Multi-Temporal Remote Sensing Cloud Removal},
  author = {Zhang, Yifan and Chen, Qian and Liu, Yi and Li, Wengen and Guan, Jihong},
  journal = {arXiv preprint arXiv:2602.00536},
  year = {2026},
  url = {https://arxiv.org/abs/2602.00536},
  note = {Yi Liu is a co-first author.}
}

CVPR
Effective Cloud Removal for Remote Sensing Images by an Improved Mean-Reverting Denoising Model with Elucidated Design Space

Yi Liu, Wengen Li, Jihong Guan, and 2 more authors

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2025

Abs Bib HTML PDF Code

Cloud removal (CR) remains a challenging task in remote sensing image processing. Although diffusion models (DM) exhibit strong generative capabilities, their direct applications to CR are suboptimal, as they generate cloudless images from random noise, ignoring inherent information in cloudy inputs. To overcome this drawback, we develop a new CR model EMRDM based on mean-reverting diffusion models (MRDMs) to establish a direct diffusion process between cloudy and cloudless images. Compared to current MRDMs, EMRDM offers a modular framework with updatable modules and an elucidated design space, based on a reformulated forward process and a new ordinary differential equation (ODE)-based backward process. Leveraging our framework, we redesign key MRDM modules to boost CR performance, including restructuring the denoiser via a preconditioning technique, reorganizing the training process, and improving the sampling process by introducing deterministic and stochastic samplers. To achieve multi-temporal CR, we further develop a denoising network for simultaneously denoising sequential images. Experiments on mono-temporal and multi-temporal datasets demonstrate the superior performance of EMRDM. Our code is available at https://github.com/Ly403/EMRDM.
@inproceedings{liu2025effective, title = {Effective Cloud Removal for Remote Sensing Images by an Improved Mean-Reverting Denoising Model with Elucidated Design Space}, author = {Liu, Yi and Li, Wengen and Guan, Jihong and Zhou, Shuigeng and Zhang, Yichao}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, year = {2025}, pages = {17851--17861}, url = {https://arxiv.org/pdf/2503.23717}, }