Career
Joined ByteDance as a Byte Intern.
Working on video generation and reward modeling.
Open to 2027 full-time opportunities
Tianxiao Chen
Research in video generation and multimodal learning.
I'm a master's student at Zhejiang University and a Byte Intern at ByteDance, working on AIGC, video generation, multimodal intelligence, and reward modeling.
CurrentlyVideo generation · Multimodal learning · Reward modeling
5Research works
2Industry labs
2027Expected graduation
Latest updates
News
Recent publications, research milestones, and academic updates.
Publication
QTac was accepted to IROS 2026.
Our work on quantitative robotic grasping with tactile sensing.
Career
Joined vivo BlueImage Lab as a research intern.
Worked on video generation and developed VIPER during the internship.
Publication
MIMIC was accepted to ICLR 2026.
Mask-injected manipulation video generation with interaction control.
Publication
RePrompt was accepted to IEEE TCSVT.
Retrieval-enhanced visual prompt learning for few-shot classification.
Education
Started my master's journey at Zhejiang University.
Joined the College of Control Science and Engineering as a ZJUer.
Education
Graduated from Zhejiang University of Technology.
Recognized as an Outstanding Graduate of Zhejiang Province.
Selected publications
Selected publications
Selected work across video generation, robotics, vision-language learning, and 3D reconstruction.
01MIMIC: Mask-Injected Manipulation Video Generation with Interaction Control
Controllable manipulation video generation that learns interaction dynamics from reference examples.
02VIPER: Visual In-Context Physics Reasoning for Physically Plausible Video Generation
Visual in-context reasoning that transfers physical behavior from reference videos to improve plausibility.
03QTac: Quantitative Robotic Grasping with Tactile Sensing
Quantitative tactile reasoning for robotic grasping, connecting language-level goals with contact feedback.
04Retrieval-Enhanced Visual Prompt Learning for Few-shot Classification
A retrieval-enhanced visual prompt framework for adapting vision-language models in few-shot settings.
052L3: Lifting Imperfect Generated 2D Images into Accurate 3D
A 2D-to-3D lifting framework designed to recover accurate geometry from imperfect generated images.
Background
Education & experience
Academic training in control and engineering, together with industry research experience in video generation.
Education
Zhejiang University of Technology
Jianxing Honors College · Hangzhou, China
Undergraduate education
Top 5%
Education
Zhejiang University
College of Control Science and Engineering · Hangzhou, China
Master's student
Advisor: Qi Ye
Industry
vivo
Research Intern · Hangzhou, China
Video generation
Mentor: Pengtao Jiang
Industry
ByteDance
Byte Intern · Hangzhou, China
Video generation · Reward models · Reinforcement learning
Mentor: Pengqi TuContact
Get in touch
I'm looking for full-time research and engineering opportunities starting in 2027, and I'm open to research discussions and collaboration.