Ivan

Ivan

About

Just got the master degree in civil engineering from National Taiwan University in July, 2019. Professional with image processing, image-based hydrologic remote monitoring. A machine learning enthusiastic and experienced in programming with Python using Tensorflow and Keras.
Landscape Photographer.

Quick Facts

  • Programming Languages: Python, MATLAB, SQL.
  • GUI: PyQt

Machine Learning

  • Frameworks: Tensorflow, Keras.
  • NN: Logistic Regression, CNN, RNN, Deep Auto-encoder, Genetic Algorithm.
  • Model: R-CNN based object detection models, YOLO v1~v3.
  • Image Processing: OpenCV, MATLAB.
  • Version Control: Git.

Data Science

  • Library: numpy, pandas, scipy, matplotlib.
  • Statistic: ANOVA, Regression, Uncertainty analysis, Reliability analysis, Time series, Kriging method.

Civil Engineering

  • Knowledge: Fluid Mechanics, Open Channel Flow, Hydrology.
  • Numerical Model: HEC-HMS, HEC-RAS, SWMM.
  • GIS: Arc-GIS, QGIS, UAV, Agi PhotoScan 3D Terrain Reconstruction.
  • Construction Management: MS Project.
  • Technical Drawing: AutoCAD.

Personally

  • Photography.
  • Adobe Photoshop, Lightroom.

Projects

Mask R-CNN for Cookies Detection and Segmentation

GitHub repo | Post
This is a tiny project to use Mask R-CNN for detecting two brands of cookies “Lays” and “Doritos”. Most of the code is based on the implementation of Mask R-CNN by matterport on Python 3, Keras, and TensorFlow. What we modified is changing the backbone network from ResNet-101 to ResNet-50 and the batch size from 2 to 1 image. This setting will use 97~98% memory of NVIDIA RTX2060 6GB.

pyPIV - A Particle Image Velocimetry GUI toolkit

GitHub repo
This is the project dealing with Particle Image Velocimetry based on two algorithm:

  1. Direct Cross Correlation (DCC)
  2. Convolutional Neural Network (CNN)

The GUI files built with PyQt helps user to modify the parameters in the algorithm more easily.

Note: CNN method is not open to public. See the section Master’s Thesis below.

Master’s Thesis

< The Application of Convolutional Neural Network on Large-Scale Particle Image Velocimetry >
The research made an effort on improving the river measurments results on the field by conventional LSPIV technique. We implement a tiny self-build CNN model to replace the direct cross-correlation algorithm (DCC) in the conventional one. We found out that CNN-based LSPIV can keep the decrease of performance below 1% which is 18% and much more unsteady by DCC-based under the noise of illumination. The method we proposed is robust and can give an excellent velocity field with stability and accuracy then DCC gives.

歡迎來到我基於GitHub pages建置的部落格，而主題樣式則採用Michael Rose所發表的Jekyll主題”Minimal Mistakes“。
Welcome to my blog built on GitHub pages. The theme is Minimal Mistakes designed, developed, and maintained by Michael Rose.

本部落格主要語言為中文、英文。
Primary languages here: Chinese, English.