Zhao-Yang Zhuang (莊昭陽)

Experience with computer vision, deep learning. I am very passionate about design a novel DL model architecture for real world problems and hope to participate in more types of applications.

ML engineer Data scientist

[email protected] https://www.linkedin.com/in/rogan-zhuang/

https://github.com/ZhuangRogan

Educations

National Taiwan University

Civil Engineering, Computer-Aided Engineering Group (M.S. degree)

(2018 ~ 2020)

Master Thesis:

Deep-learning method assisted crane for sway prediction: Recurrent Kalman Network.

Work Experiences

(Automation Services Co., LTD.) (AI/CV Engineer)

(Feb, 2021 - Oct, 2021)

The main developer of AI/CV algorithm.
Compiled the AI/CV part of automatic fabric inspection machine alone from scratch.

Research Experiences

Deep-learning method assisted crane for sway prediction : Recurrent Kalman Network

(Dec, 2019 - Jun, 2020)

Robotic LAB, NTU & 祐彬 Construction Co., Ltd

Predicting the future position of a payload by RKN to assist in the automation of crane.
The state of the payload system can be inferred by the combined information between the image pixels.

Key achievement: Assists the automation of the crane in a new way, and does not require any physical parameters but only image data of the payload. RKN, the new time-series forecasting architecture will be 10x better than LSTM.

Tensorflow(Keras) OpenCV Python C/C++ Deep Learning

The second generation of automatic fabric inspection machine

(Feb, 2021 - Apr, 2021)

Automation Services Co., LTD.

Detecting the defects on the textile by deep learning that efficiency is at least ten times of manual visual inspection.
Calculate the area and location of defects by semantic segmentation.
The prototype machine can detect the width of about two meters with eight cameras.

C# Deep Learning

The third generation of automatic fabric inspection machine

(May, 2021 - Oct, 2021)

Automation Services Co., LTD.

Based on the second-generation inspection machine, all of the features are upgraded, such as image grabbing system, camera calibration system, deep learning framework, thread optimization, etc.
This mass production machine is online.
Not implemented using Tensorflow or Pytorch.

Key achievement: 100% detection of defects required by customers, 36x8 frames per second detection speed.

C# Deep Learning Cameras System Multithread Programming

Side Projects

Image automatic labeling: CNN-base object tracking by Siam-Net. [labeling 90x faster than manual]
Document layout analyzing: locate file elements (titles, text, pictures, etc.) by image segmentation.
HousePrice-prediction: Predict the HousePrice in Feature Engineering & Advanced Regression. [Kaggle top 0.04%]
Flowers classification: Distributed training a CNN to classify 104 kinds of flower pictures. [Kaggle top 0.09%]