About Me
Hi, my name’s Yi Te and I’m a cutting-edge algorithm developer, especially for computer vision. I have 3+ years working experience on vision algorithm and related programming. As well as a good consultant in the teamwork. I’m most skill in C++ and Python
Projects
LucidPix 3D Photo Rendering Optimization
RD & Lead
There are a variety of processes to generate a 3D photo, including depth generation, inpainting, sementic segmentation, mesh generation, and rasterization. Each process has its issues, and is related to other processes. Our goal is to provide user a friendly 3D Photo via modern computer vision and computer graphics method.
3D Contour Trajectories Reconstruction through Multiple Views
https://yitech.github.io/portfolio/3D-Contour-Trajectories-Reconstruction-through-Multiple-Views/Nominated by ASM 2020 innovative project, RD & Lead
The camera pictures the wires from different sides. The project goal is to reconstruct the 3D wires’ trajectories though different wires’ image information. In this project, I not only work on the algorithm, but also participant the whole AOI process. The problem is regarded as an “innovation” because it’s not just a 3D reconstrucion problem. The difficulties include multiple contours overlap in 2D and not observe 3D trajectories directly. Eventually, we came out a sequence of brilliant skills to shoot these issues.
Agent-Based Job-Shop Scheduling Optimization
RD
This is a popular topic for a smart factory. The agents are able to transfer the material into different processes. Try to come out a general algorithm to optimize the scheduling of agents. In this project, we use Directed Acyclic Graph to describe the process order of scheduling, and use a min-flow and Monto Carlo tree search to find a near-optimal scheduling in the real application. Besides, we also have some contribution of Reinforcement Learning in this project.
Field of View Path Planning Optimization
https://yitech.github.io/portfolio/Field-of-View-Path-Planning-Optimization/RD
The Feild of View(FoV) is described as the extent of the observable world that is seen at any given moment. The project target is going to optimize the time cost of inspection, such that the camera can picture all the inspector in the FoVs. In the project, I encode the solution of a sequence of inspected objects and apply Genetic Optimization to improve the FoV performance.
Experience
I am responsible for 3D computer vision analysis and audio analysis. For 3D computer vision, I design the potential features of 3D photo application, prove of concept and implementation. For audio analysis, I mainly work on audio semantic segmentation. It is a diversified and brain-storming job.
This position is responsible for developing the solutions of computer vision. Overall, I take about 20% on requirement engineering, 50% on paper work, including paper study, proof of concept and deriving the mathematical formula. The remaining 30% is used for architect and implementation. I think this position needs good communication and sense of mathematics. A good communication can pool different kind of senses together to find the best solution, and a good sense of mathematics provide an efficient diagnosis while facing issues.
Applying the deep learning and machine learning algorithm, develop the foresighted AOI system. I learnt image processing, image recognition and integrate AI model into AOI system.
Education
National Chiao Tung University
Master of Statistics
Sept 2014 – June 2016
National Chiao Tung University is a public research university located in Hsinchu, Taiwan. It is one of Taiwan's leading universities.
During my time of NCTU I specialize in Probability and Statistical Computing, which is a branch of Computer Science and Statistics. I have good understanding of fundamental science and can apply the cutting-edge paper efficiently.
My thesis is a study of Dynamic Weighted Monto Carlo Markov Chain(MCMC). MCMC is a widely used statistical computing method. However, the inefficient convergence property is notorious in application. We purpose a weighted scheme of MCMC to enhance the convergence and provide a numerical and mathematical proof of the efficientcy.
National University of Kaohsiung
BSc Applied Mathematics
Sept 2010 – June 2014
National University of Kaohsiung is a national university located in Kaohsiung, Taiwan.
I got into mathematics in these four years. In this period, I have studied much different kind of math, both abstract and applied math included. It’s the seedbed of the engineer for me.
Skills
- deep learning solutions, including feature detection, feature extraction, object detection, object recognition, object tracking and image segmentation.
- image processing algorithms, including image matching, space/time domain analysis, morphology, ranging and depth, etc.
- point cloud processing, including feature estimation, point cloud registration, surface reconstruction, 3D perception.
- statistical optimization, including EM algorithm, Monto Carlo method, reinforcement learning, Bayesian optimization, genetic algorithm.
- computer graphics, including rasterization, digital geometry processing, camera projection, color radiometry.
- data cleaning, data preprocessing, and data visualization.