莊承翰
The devil is in the detail.
Video and Image Algorithm, Deep learning, Software development
Summary
- Over 8 years of experience in deep learning and more than 10 years of experience in computer vision/image processing.
- Capable of independently or collaboratively building AI (image/video) products from scratch with various teams.
- Analyzing, and improving deep learning systems.
Kaggle
- Top 1% ( 16 / 1559 ) Sartorius - Cell Instance Segmentation (2021/12~2021/12)
- Top 4% ( 85 / 2160 ) Human Protein Atlas Image Classification (2018/11~2019/1)
profile
Skills
Software
- C++
- Python
Library / Framework
- OpenCV
- PyTorch
- LibTorch
DL / ML
- Classification / Regression
- One stage / Two stage Object Detection
- Inbalance Data Handle
- Domain Adaptation
- Semi/Self Supervised
- Data Augmentation
- Mislabel Data Detection
- New Class Detection
- Abnormal Detection
Experience
Delta Electronics, Principal Algorithm Engineer, May 2022 ~ Present
[ Primary Job Responsibilities ]
- Development and implementation of DL-related technologies for industrial inspection applications.
- Increase the efficiency of team members across various projects by SOP and tools.
[ Projects ]
- DIP Manufacture Defect Detection (Internal Factory)
- Human replacement rate 50-70% (depending on different product conditions).
- Plastic Fan Defect Detection
[ Concept Verification ]
- Analog Meter Recognition
[ Skill ]
- Deep Learning (Pytorch)
- Transitional CV
- C++ / Python
Delta Electronics, Senior Algorithm Engineer, Jul 2017 ~ May 2022
[ Primary Job Responsibilities ]
- Development and implementation of DL-related technologies for industrial inspection applications.
- Algorithm system development and maintenance.
[ Projects ]
- SMT Manufacture Defect Detection (Internal Factory)
- Human replacement rate 50-70% (depending on different product conditions).
- PCB Ceramic Basic Defect Detection (External Clients)
- Establishment of acceptance standards.
- Human replacement rate 60-80% (depending on different product conditions).
- Adoption of Domain Adaptation-related technologies to reduce data collection and accelerate model deployment time for new products.
- TFT-LCD Defect Detection (Inherited Project) (External Clients)
- Accuracy rate 85-95% (depending on different product lines).
(Improved accuracy rate by 2-8% compared to previous. )
[ Concept Verification ]
- Metal Scratch Detection
- Patent Number: I791970
- Analog Meter Recognition
[ Skill ]
- Deep Learning (Pytorch)
- Transitional CV
- C++ / Python
ADE Technology, Algorithm Engineer, Apr 2016 ~ Jul 2017
[Projects]
- License Plate Recognition System
- Algorithms development
- UI
- Pedestrian Detection
- Algorithms development
- Thermal Image Enhancement
- Transitional CV
- C++ / Python
Genius Vision Digital, Algorithm Engineer, Oct 2014 ~ Feb 2016
[Projects]
- License Plate Algorithms
- People Counting Algorithms
- System Debug and Maintain
- Significantly improve the executive performance of people counting algorithms (Reduce the CPU loading by 50%)
- Transitional CV
- C++
Education
National Chin-Yi University of Technology, 2011 ~ 2013
Master of Computer Science and Information Engineering
National Chin-Yi University of Technology, 2007 ~ 2011
Bachelor of Computer Science and Information Engineering
