I am currently an AI engineer at CyberLink, where I specialize in computer vision, deep learning, and generative AI. In 2023, I earned my M.S. degree from the Graduate Institute of Communication Engineering (GICE) at National Taiwan University, focusing on data science under the guidance of Professor Pei-Yuan Wu. With over four years of research experience, my interests span Deep Learning, Computer Vision, Machine Learning, and Audio Processing.
- Developed an approach to project future vehicle trajectories onto current front-view images, serving as ground truth for the 3D vehicle future trajectory prediction model. This approach utilizes data collected by vehicles equipped with RTK and front-view cameras, enabling model training with minimal human intervention.
- Developed and optimized a 3D vehicle future trajectory prediction model, enhancing performance through empirical experiments focused on dataset improvement, warping method, and data augmentation. Conducted model quantization for deployment on embedded devices.
- Deployed the quantized 3D vehicle future trajectory prediction model on the Qualcomm 8295 platform by converting the Python program to C++. Optimized model efficiency to achieve 200 FPS, and the quantization error (L1 error) was reduced from 4.0 to 0.15 ( ↓266%).
- Collaborated with team members to integrate the quantized 3D vehicle future trajectory prediction model into the AR navigation algorithm, enabling on-vehicle testing. This enhancement significantly improved performance in challenging driving scenarios, such as navigating roundabouts and sharp turns.
- Designed an intersection depth estimation auto-labeling algorithm using projected RTK trajectories.
- Implemented general frameworks to support various deep learning tasks including recognition, segmentation, and object detection.
- Enhanced the existing wound classification model, improving accuracy by approximately 3%.
- Deployed models as a service using Docker on GCP.
- Developed Gaze Estimation Model using a self-supervised learning technique, successfully reducing the error rate by 90%.
- Designed a feature matching algorithm to accurately detect torsional eye rotation.
- Advisor: Prof. Yu-Chiang Frank Wang
- Designed and grade homework sets and final project
- Motivated students during T.A. office hours
- Advisor: Prof. Pei-Yuan Wu
- Designed the programming exercises
- Advisor: Prof. Jian-Jiun Ding
- Graded homework sets
- Advisor: Prof. Pei-Yuan Wu
- Designed and graded the theoretical and programming homework sets
[ICIP 2022] CTGAN: Cloud Transformer Generative Adversarial Network
Github link
2022 T-Brain competition - Lung Adenocarcinoma Pathological image segmentation
Private Leaderboard: 2/307, Top1%
Github link
2022 AIdea competition - Crops Status Monitoring by Image Recognition
Private Leaderboard: 3/428, Top1%
Github link