Tianhao (Terry) Fu

I’m currently an undergraduate student in Engineering Science 2T9 at UTSG. My major interest is in the field of Computer Vision, which include Autonomous Racing and Medical Image Processing. I am also a full stack developer who mainly focus on Next.js, SpringBoot, and MySQL.

I initially worked as a full stack developer since middle school in Shanghai. I developed the website and schedule system for Intelligent International Education as a part-time co-op job in 2023. My research career started in high school at Villanova College. Later, I did an internship in MIP at Vector Institute.

Earlier in 2019, I founded Project Neura and held tens of projects with other members. Noticeably, we developed LEADS as a collaborative project with the VeC team, which is a club in my high school that aims to compete in the University of Waterloo EV Challenge. I’m still active in the VeC Steering Council even though I have left high school.

Also, a big fan in racing and sport cars who have heavily invested in Ferrari and Porsche.

Publications

LEADS: Lightweight Embedded Assisted Driving System

PROJECT + PAPER + REPO

With the rapid development of electric vehicles, formula races that face high school and university students have become more popular than ever as the threshold for design and manufacturing has been lowered. In many cases, we see teams inspired by or directly using toolkits and technologies inherited from standardized commercial vehicles. These architectures are usually overly complicated for amateur applications like the races. In order to improve the efficiency and simplify the development of instrumentation, control, and analysis systems, we propose LEADS (Lightweight Embedded Assisted Driving System), a dedicated solution for such scenarios.

A Multi-task Model for Joint Segmentation and Classification of Pancreatic Cancer

REPO

Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive malignancy with poor prognosis, underscoring the critical need for early and accurate diagnosis.We present a novel two-stage pipeline for PDAC detection from contrast-enhanced computed tomography (CECT) scans, leveraging the nnU-Net framework for segmentation and a high-resolution multitask convolutional neural network (CNN) for joint lesion segmentation and classification. The first stage employs a low-resolution nnU-Net to segment the pancreas region, while the second stage refines segmentation and performs lesion classification. The methods achieved an AUROC of 0.9833 and an AP of 0.8011 on the PANORAMA challenge dataset.

Towards Robust Instrument Segmentation with Data Augmentation

REPO

Accurate and robust segmentation of surgical instruments is critical for effective analysis of surgery videos. In this paper, we present our solution to the MICCAI 2024 SegSTRONG-C challenge, focusing on enhancing segmentation performance under challenging conditions. We introduce a data augmentation technique that simulates the presence of surgical smoke. Combined with AutoAugment, we train a U-Net model to automatically segment instruments and the model achieves mean Dice Similarity Coefficient scores of 0.87, 0.43, and 0.92 across the domains of smoke, low brightness, and blood, respectively. The code is publicly available at https://github.com/ProjectNeura/SegSTRONGC.

AQuiz

论文 + 仓库

AQuiz 是为 ArRow 设计的推荐算法，前身为原 XFCRC。

A Deep Learning Model for Accurate and Robust Internet Traffic Classification

PAPER + REPO

In this paper, we proposed a deep learning model which achieves progress compared to LeNet-5 in the stability of Internet traffic classification.