Huimin Zeng (曾慧敏)

I am currently a second-year Ph.D. student from SmileLab at Northeastern University (NEU), supervised by Prof. Raymond Fu. Before that, I received my Master's degree from the University of Science and Technology of China (USTC) in 2024, supervised by Prof. Zhiwei Xiong. I received my Bachelor's degree at Ocean University of China (OUC) in 2021.

Email  /  LinkedIn  /  WeChat  /  GitHub  /  Google Scholar

News

❗️ I am currently looking for research internship opportunities for Summer 2026. Thanks for any referrals or reaching out if there's a potential fit!

[Dec. 2025]  Passed my PhD qualifying exam, thanks to my advisor and committee members! Cheers! 🎉

[Nov. 2025]  One paper for 3D super-resolution is accepted to AAAI 2026.

[Sep. 2024]  Begin my journey at Northeastern University.

Research

My research focuses on adaptive and interpretable computational photography, with a particular emphasis on 3D scene reconstruction, MLLMs, and low-level vision. I'm also open to discussions and collaborations. 🤝 Feel free to reach out!

Arbitrary-Scale 3D Gaussian Super-Resolution

We make the first attempt to achieve 3D super-resolution of arbitrary scale factors with a single 3DGS model, providing a unified and efficient solution for flexible, high-resolution novel view synthesis.

Plug-and-Play Versatile Compressed Video Enhancement

We introduce a versatile quality enhancement framework that leverages bitstream information to adaptively enhance videos of different compression levels and assists various downstream vision tasks.

All-in-One Image Compression and Restoration

We design the first unified pipeline for all-in-one image compression and restoration, which equips the neural image codec with the restoration capability and improves its generalization ability against various degradations.

MarineInst: A Foundation Model for Marine Image Analysis with Instance Visual Description

We propose MarineInst, a powerful and flexible marine foundation model, which could perform the instance visual description task in an automatic or interactive manner.

Region-Aware Portrait Retouching with Sparse Interactive Guidance

We explore the interactive portrait retouching task, allowing the retouching model to effectively capture users’ intents and deliver visually satisfying retouched results.

Feature Decoupling-Recycling Network for Fast Interactive Segmentation

We propose a feature decoupling-recycling scheme, which decouples image semantics extraction and user interaction processing, while recycling extracted feature maps to improve efficiency.

Design and source code from Jon Barron's website