Gaussian-Based Colorization and Super-Resolution for 3D Reconstruction

06.2024-12.2024   Github

Outline

Pioneered an innovative system for colorizing and three-dimensionally reconstructing monochrome historical films and documentaries, enhancing the preservation and visualization of archival footage.

Key Responsibility

Engineered a real-time colorization pipeline integrating segmented optical flow and the DeOldify algorithm, coupled with ColMap for feature extraction.

Developed an end-to-end 3D reconstruction framework utilizing Gaussian Splatting, enabling immersive visualization of colorized historical content.

\(BMS^3\): Bayesian Modeling Based SwinUNet Segmentation on Self-distillation Architecture

05.2024-05.2025

Outline

Developed a novel approach for medical image segmentation enhancing domain invariance and generalization.

Key Responsibility

Integrated Bayesian modeling with Swin Transformer-based U-Net architecture and implemented self-distillation mechanism, conducting experiments on multiple prostate MRI datasets.

Achievement

Accepted for Oral in ICONIP 2025

ESP32-based Real-Time IV Drip Monitoring and Alert Platform

11.2023-04.2024   GITHUB    DEMO

Outline

Developed an innovative IoT-based system for real-time monitoring and control of intravenous drips in hospital settings.

Key Responsibility

Designed and implemented an integrated system using ESP32, incorporating drop sensors for real-time monitoring, servo motors for flow control, wireless communication for alert transmission, and a centralized nurse terminal as monitor for multiple IV stations.

Mutual Information Calculation on Different Appearances

11.2023-04.2024

Outline

Conducted research on applying mutual information (MI) to assess similarity between images, particularly focusing on comparing appearances of different individuals.

Key Responsibility

Implemented and analyzed mutual information, entropy, and information gain algorithms for image matching and similarity assessment, including pre-processing techniques, probability density function calculations, and performance evaluations across various image scenarios.

U-Net Conditional GAN-Based Data Augmentation in Classification Problem with Low Data Resource

10.2023-12.2023

Outline

Modified an innovative data augmentation technique using conditional Generative Adversarial Networks (cGANs) to address low data resource challenges in medical image classification.

Key Responsibility

Designed and implemented a U-Net based cGAN architecture for generating synthetic medical images, integrating it with classification models (ResNet, DenseNet) to enhance performance on datasets including ChestXray8, LiTS, NCT-CRC-HE-100K, and BreastUltra.

Precision Area Control and Line Crossing Alerts based on YOLOv8

11.2023-12.2023

Outline

Developed an advanced real-time detection system for traffic monitoring and human tracking applications on certain area and lines.

Key Responsibility

Implemented a YOLOv8-based detection system with custom zone counting and cross line detection functionalities, adapting and fine-tuning the COCO-trained model to optimize performance for specific traffic and human detection requirements.

HuBMAP + HPA - Hacking the Human Body (Kaggle Competition)

07.2022-10.2022

Outline

Participated in a Kaggle competition focused on identifying and segmenting functional tissue units (FTUs) across five human organs using tissue section images.

Key Responsibility

Developed a semantic segmentation model using ASPP and FPN for feature extraction, implementing model fusion techniques to enhance accuracy and reduce complexity, achieving a public score of 0.79 on Kaggle.

Achievements

Won a bronze medal in the competition.

OpenGL My World: Interactive 3D Environment

04.2024-05.2024    GITHUB

Key Features

Developed an immersive 3D world using OpenGL with dynamic object tracking and interaction

Implemented first-person movement, diverse textures, and detailed building structures

Technical Achievements

Optimized performance for smooth rendering and interaction

Implemented collision detection, random object generation, and gravitational physics

Integrated skybox for enhanced visual immersion

Developed robust user input handling and debugging systems

BCI Signal Processing

12.2023-02.2024

Main

Utilize transfer learning on the “efficientnet” to classify six kinds of harmful brain activities based on EEG signal, and use both the semantic data signal and SWIN transformer to make the prediction.

Add new noise reduction methods to deal with the accident pulse, and finish the competition of HMS on Kaggle and the database obtains a high score; Works on this project will continue.

Compiler Construction Development based on C

09.2023-01.2024    GITHUB

Main

Constructed a new compiler in C in 3 phases: lexical analysis, syntax analysis and semantic analysis. Implemented a C-based compiler with a DFA lexer, AVL tree symbol table, a parsing table for shift/reduce operations, and Hindley-Milner type checking with β-reduction.

Non-chordal Music Generation

10.2022-11.2022

Main

Designed and implemented a non-chordal music generation system using Bi-LSTM. Provided two ways for generating the non-chordal music, Auto-generation and Continuation Generation. Employed temperature sampling and designed lexical lists to map the corpus for digital storage.

Network Communication Software

06.2023    GITHUB

Main

Developed a network communication software using C++ with QT for the GUI, featuring a multi-threaded server and client for data transmission, and MySQL for data storage.

Software of Film Analysis, Prediction and Auto-arrangement for Theatre

05.2023-06.2023

Main

Developed a software using QT and MySQL for film analysis, prediction, and auto-scheduling, featuring deep learning for data analysis and multiple visualization methods.

Designed Database for Medical Donation and Web Processing

10.2022-12.2022    GITHUB

Main

Created a web-based donation platform that aims to facilitate donations to poverty and disaster-affected areas worldwide.

Designed a relational database for a web-based donation platform using MySQL, focusing on entities such as users, beneficiaries, donations, and comments, with a front-end implemented in Bootstrap.

Webpage and Database Design for Book Purchase and Exchange Store

05.2022-06.2022    GITHUB

Main

Developed a comprehensive book platform allowing users to purchase and exchange books, with the entire system self-developed.

Library Lending System

05.2022-06.2022    GITHUB

Main

Developed a Java-based library book lending system with a command-line interface, featuring user management, book tracking, and error handling, structured with interfaces and abstract classes.