CV

Ph.D. in Computer Science

Khoury College of Computer Sciences, Northeastern University (NEU)

• Advisor: Prof. Huaizu Jiang

M.S. in Computer Science

Viterbi School of Engineering, University of Southern California (USC)

• GPA 4.0/4.0

B.Eng. in Computer Science and Technology

School of Computer Science & Technology (SCST), Huazhong University of Science and Technology (HUST)

• GPA 3.99/4.0, Rank 2/363
• Selected by SCST for a special class for the most promising students

Research Assistant

Visual Intelligence Lab, Northeastern University — Advisor Prof. Huaizu Jiang

• Introduced SNAP, the first unified segmentation model capable of handling diverse point cloud domains (indoor, outdoor, aerial) while supporting versatile prompt types such as spatial points (clicks) and text for flexible object segmentation.
• Introduced Struct2D, a perception-guided prompting framework that unlocks spatial reasoning in Multimodal Large Language Models (MLLMs) using structured 2D representations.

Research Assistant

TACO Group, Texas A&M University — Advisor Prof. Zhengzhong Tu

• Presented a novel perspective for protecting personal images from malicious editing, focusing on making biometric features unrecognizable post editing.
• Developed FaceLock, an algorithm incorporating facial recognition models and feature embedding penalties to effectively protect against diffusion-based image editing.
• Conducted a crucial analysis of the quantitative evaluation metrics commonly used in image editing tasks, exposing their vulnerabilities and highlighting the potential for manipulation to achieve deceptive results.

Research Assistant

Embedded and Pervasive Computing Lab, HUST — Advisor Prof. Xianzhi Li

• Modified a few-shot learning framework for 3D Instance Segmentation (3DIS) to address the high cost of collecting sufficient annotated point clouds.
• Utilized a Transformer Decoder to generate differentiated kernels to perform instance-wise dynamic convolution.
• Implemented the model using Python and PyTorch, improving mean Average Precision (mAP) by 3.2 percent on the ScanNet V2 dataset.

Assistant Algorithm Engineer

Research & Development Group (RDG), iFLYTEK

• Worked on a 3D Instance Segmentation project combining clustering- and Transformer-based methods, achieving 0.796 on the ScanNet V2 AP50 benchmark (surpassing the prior state-of-the-art of 0.787).
• Modified the indoor scene instance segmentation model to improve performance on outdoor scene datasets.

• Outstanding Graduates of HUST

• Merit Student of HUST

• Merit Student of HUST

• China National Scholarship (the highest national-wide scholarship for undergraduate students in China)
• Outstanding Undergraduates in Term of Academic Performance (the greatest honor for undergraduates in HUST)
• Merit Student of HUST

Single-Source Domain Generalization Project

• Led a team of 4 through a semester-long research project applying the Segment Anything Model (SAM) to Single-Source Domain Generalization for Medical Image Segmentation, resulting in a first-author preprint.
• Set the technical direction (dual-stage fine-tuning paradigm, mask-filtering module) and coordinated task allocation, manuscript drafting, and poster creation across the team.
• Achieved state-of-the-art results on the Prostate dataset, 8% above the prior best.

Stock Management Website

• Designed and implemented the frontend using HTML5, Bootstrap, and Angular; engineered the backend with Node.js, including server-side logic, database integration, and API handling.
• Managed a cloud-based MongoDB database and deployed the site on Google Cloud Platform (GCP).

Stock Management Android App

• Developed a stock management Android application using Android Studio, refactoring the Node.js backend for mobile use.
• Utilized Volley and Picasso to handle multiple HTTP requests.

Jigsaw Puzzle Project

• Led a group of 4 to develop a jigsaw puzzle game-playing website using JavaScript and the Paper.js graphics framework.
• Designed a novel magnetic mode for fun-seeking users, deployed the site on a cloud server, and adapted it for different PC and mobile devices.