RONGMAN XU

University of Edinburgh

BSc Hons Computer Science and Mathematics

First-year average: 90.83 / 100 (A1)

Beijing No.171 High School

Junior High and Senior High

A-level: A* (Maths, Further Maths), A (Physics, Econ)

Clipart Animation Project, School of Informatics, University of Edinburgh

Supervised by Dr. Changjian Li, University of Edinburgh

• Researched clipart-style animation, focusing on view-consistent motion and sparse keyframe interpolation.
• Evaluated three pipelines: (1) direct video generation, (2) text-to-multiview synthesis with interpolation, and (3) skeleton-guided motion transition.
• Optimized animation codebase to accept skeleton-based input; implemented motion control by learning framewise keypoint offsets under structural constraints.
• Fine-tuned diffusion models for 2D clipart style.
• Worked with PhD mentors to address challenges in non-human character consistency, skeleton topology, and artifact reduction.

Semantic Occupancy for Autonomous Driving

Supervised by Dr. Hao Zhao, Tsinghua University

• Collaborated with researchers at the Institute of Automation, Chinese Academy of Sciences, strengthening teamwork and coordination skills.
• Led the implementation and optimization of GaussianFormer, an object-centric autonomous driving system based on Gaussian distributions, utilizing the LightwheelOcc dataset.
• Improved evaluation metrics and enhanced dataset preprocessing pipelines to address data quality challenges.
• Gained hands-on experience in semantic occupancy estimation and autonomous driving systems.
• Project concluded due to unresolved dataset quality issues from the third-party provider, highlighting the importance of robust data pipelines in AI research.

Durhack hackathon 2024

Developed a game called "Eco Hero" with ZhongYi Yan and Neil Armstrong

• Eco Hero is a mini-game where players can clean beaches and city streets, collecting litter and restoring beauty to the environment. By engaging players in fun gameplay, we aim to instill a sense of responsibility towards our planet and inspire real-world action.
• Devpost link

Skin Cancer Image Classification

Personal Project

• GitHub Repo
• Independently completed a paper titled "Image Classification of Skin Cancer Using Deep Neural Networks with Scaling Laws", published as the first author in the journal Inter national Journal of Computer Science and Information Technology.
• This study examines how dataset size influences image classification performance across eight networks using varying proportions of the HAM10000 dataset. The findings un derscore the critical role of dataset scale and model complexity in enhancing skin cancer classification, aiming to alleviate healthcare workload.
▶️ Try it on Hugging Face

GEF Strengthening Marine Protected Areas in SE China

Project Internship Assistant

• Utilized Python and Excel to effectively calculate data including co-financing status, gen der ratio, area of protected areas, and volunteer numbers.
• Organized and coordinated a large-scale training session and the Terminal Evaluation Kick-off Meeting, managing travel arrangements and schedules to ensure smooth project execution.
• Drafted press releases to support project promotional and outreach events.
• Coordinated project bids and successfully produced 20 bid documents.
• Provided translation support for documents between Chinese and English.