Zhicheng Sun (孙至诚)
I am a final-year Ph.D. candidate focusing on computer vision and machine learning at Peking University,
where my advisor is Prof. Yadong Mu.
Before that, I obtained my B.S. degree in Computer Science from Peking University.
Email: sunzc [at] pku [dot] edu [dot] cn
CV
 / 
Google Scholar
 / 
LinkedIn  / 
Github  / 
Twitter
|
|
Research
My research spans Generative Models and Continual Learning. My short-term goal is to improve generative models from the continual learning perspective (e.g. efficiency, knowledge sharing). My long-term goal is to enable generative models to learn continually (through new paradigms such as flow-based models and equilibrium models).
|
|
Pyramidal Flow Matching for Efficient Video Generative Modeling
Yang Jin,
Zhicheng Sun,
Ningyuan Li,
Kun Xu,
Kun Xu,
Hao Jiang,
Nan Zhuang,
Quzhe Huang,
Yang Song,
Yadong Mu,
Zhouchen Lin
Preprint, 2024
project page
/
paper
/
code
/
bibtex
We design a unified flow matching objective that jointly generates
and decompresses visual content across pyramidal representations.
|
|
Closed-Loop Long-Horizon Robotic Planning via Equilibrium Sequence Modeling
Jinghan Li,
Zhicheng Sun,
Fei Li,
Cao Sheng,
Jiazhong Yu,
Yadong Mu
Preprint, 2024
paper
/
code
/
bibtex
We propose an equilibrium model-based LLM planner that is capable of
self-refining its plan with external and internal feedback.
|
|
RectifID: Personalizing Rectified Flow with Anchored Classifier Guidance
Zhicheng Sun,
Zhenhao Yang,
Yang Jin,
Haozhe Chi,
Kun Xu,
Kun Xu,
Liwei Chen,
Hao Jiang,
Yang Song,
Kun Gai,
Yadong Mu
NeurIPS, 2024
paper
/
code
/
bibtex
We introduce a training-free approach to personalizing rectified flow,
based on a fixed-point formulation of classifier guidance.
|
|
Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization
Yang Jin,
Zhicheng Sun,
Kun Xu,
Kun Xu,
Liwei Chen,
Hao Jiang,
Quzhe Huang,
Chengru Song,
Yuliang Liu,
Di Zhang,
Yang Song,
Kun Gai,
Yadong Mu
ICML, 2024   (Oral)
project page
/
paper
/
code
/
bibtex
We present a multimodal LLM capable of both comprehending and generating videos,
based on an efficient decomposed video representation.
|
|
Exploring Orthogonality in Open World Object Detection
Zhicheng Sun,
Jinghan Li,
Yadong Mu
CVPR, 2024
paper
/
code
/
bibtex
We develop an open world object detector that exploits feature and
prediction orthogonality to continually identify unknown objects.
|
|
Countering Personalized Text-to-Image Generation with Influence Watermarks
Hanwen Liu,
Zhicheng Sun,
Yadong Mu
CVPR, 2024
paper
/
code
/
bibtex
We forge unlearnable examples against diffusion models with a robust
watermarking method that focuses on the most influential pixels.
|
|
Rewiring Neurons in Non-Stationary Environments
Zhicheng Sun,
Yadong Mu
NeurIPS, 2023   (Spotlight)
paper
/
code
/
bibtex
We propose a novel rewiring approach by permuting hidden neurons,
allowing for structural plasticity in continual reinforcement learning.
|
|
Regularizing Second-Order Influences for Continual Learning
Zhicheng Sun,
Yadong Mu,
Gang Hua
CVPR, 2023
paper
/
arXiv
/
code
/
bibtex
We identify a new class of second-order influence functions in replay-based
continual learning, and address it with a regularized selection strategy.
|
|
Patch-based Knowledge Distillation for Lifelong Person Re-Identification
Zhicheng Sun,
Yadong Mu
ACM Multimedia, 2022
paper
/
code
/
bibtex
We use adaptively-chosen patches (rather than whole images) to pilot the
forgetting-resistant distillation for lifelong person re-identification.
|
Service and Teaching
- Conference Reviewer: ICLR 2025, NeurIPS 2024, CVPR 2025, ICCV 2023, ECCV 2024, ACM Multimedia 2022-2024, AISTATS 2025, ACCV 2024, ACML 2024, WACV 2024
- Journal Reviewer: TMM 2022-2024, TNNLS 2024, TCSVT 2024, Neurocomputing 2022
- Teaching Assistant: Computer Vision and Deep Learning by Prof. Yadong Mu, Spring 2022
- Teaching Assistant: Deep Learning: Algorithm and Application by Prof. Yadong Mu, Spring 2024
|
Miscellaneous
-
After the breakthrough of AlphaGo, I became obsessed with AI bots
and held a top-5 user ranking on the AI competitive platform Botzone for a long time.
|
|