Efficient long-video understanding
Exploring efficient long-video understanding with vision-language models, with a focus on visual token compression.
Submitted to ICLR 2027
Generative vision & image editing
About me
I am a PhD student in Computer Science and Technology at Xiamen University, working in the MAC Lab with Prof. Rongrong Ji and Assoc. Prof. Yiyi Zhou.
My research focuses on efficient diffusion inversion and controllable image editing. I develop methods to map real images back to noise, reconstruct them faithfully, and edit their content with precise control. My work includes EasyInv, DeepInv, and ObjectAdd.
My recent work also explores efficient long-video understanding with vision-language models. I am currently an intern at naive AI, a startup, working on synthetic data. Previously, I studied at the University of Sydney and the University of Nottingham Ningbo China.
- Diffusion models
- Image inversion
- Controllable generation
Current research
Synthetic data at naive AI
I am currently working on synthetic data as an intern at naive AI.
Selected publications
2025–2026DeepInvPreprint2026
DeepInv: A Novel Self-supervised Learning Approach for Fast and Accurate Diffusion Inversion
arXiv:2601.01487, 2026
A self-supervised inversion solver that learns from pseudo-noise supervision with iterative, multi-scale training for fast and accurate image-to-noise mapping.
ObjectAddPattern Recognition2026
ObjectAdd: Adding Objects into Image via a Training-free Diffusion Modification Fashion
Pattern Recognition, 174, 112807, 2026
Training-free object addition with control over placement, designed to preserve the surrounding image while integrating new content.
EasyInvICML2025
EasyInv: Toward Fast and Better DDIM Inversion
International Conference on Machine Learning (ICML), 2025
An efficient DDIM inversion method that strengthens the contribution of the initial latent state to improve reconstruction without costly iterative refinement.
Earlier work
Detection of Individual Trees in UAV LiDAR Point Clouds Using a Deep Learning Framework Based on Multichannel Representation
Z. Luo, Z. Zhang, W. Li, Y. Chen, C. Wang, A. A. M. Nurunnabi, J. Li
IEEE Transactions on Geoscience and Remote Sensing, 60:1–15 · Co-first author
A Local Topological Information Aware Based Deep Learning Method for Ground Filtering from Airborne LiDAR Data
Z. Luo, Z. Zhang, W. Li, H. Lin, Y. Chen, C. Wang, J. Li
IEEE IGARSS, 2021, pp. 7728–7731
Extraction of Power Lines and Pylons from LiDAR Point Clouds Using a GCN-based Method
W. Li, Z. Zhang, Z. Luo, Z. Xiao, C. Wang, J. Li
IEEE IGARSS, 2020, pp. 2767–2770
Education
Xiamen University
PhD student, Computer Science and Technology
MAC Lab · Advisors: Rongrong Ji and Yiyi Zhou
Sep 2023 – Present
University of Sydney
Master’s degree, Information Technology
Jul 2022 – Jun 2023
University of Nottingham Ningbo China
BSc, Computer Science with Artificial Intelligence
Upper Second Class Honours
Sep 2017 – Jun 2021
Experience
naive AI
Intern · Synthetic Data
Working on synthetic data at an early-stage AI company.
Present
CATL
Algorithm Intern
Developed LISA-based algorithms for natural-language-driven recognition in industrial images.
Sep – Dec 2024
Meiya Pico
Research Intern
Researched text matching for geographic information.
Mar – Jun 2022
Research & engineering projects
Agents for information retrieval and self-evolution
Dec 2025 – PresentDeveloping modules for information retrieval, knowledge updates, and agent self-evolution.
Language models for professional document writing
Jul 2025 – PresentDeveloping agent-orchestrated document workflows with 32B, 14B, and 7B language models.
Reservoir water-level prediction
Sep 2020 – Jun 2021Research and system development using spatiotemporal graph convolution at the University of Nottingham Ningbo China.