//assets.cakeresume.com/assets/contentarea/contentarea-2d64eca8dae2a1ffb31e0fb1df887bbad072c7c19734d7d30fa14884bbcc299a.css

https://www.youtube.com/watch?v=dVLWum9r-3k&ab_channel=Che-KueiChang

Human Body Model Generation from 1D Measurements  

Software Engineer

張哲魁

Design high performance software to solve real-world problems efficiently. Deal with huge data and process the data. 
Research complicated technique obstacles and propose novel approaches.


National Taiwan University

Aemass

Computer Vision Engineer

3D Reconstruction:
1. Reconstruct the dynamic 3D model in near real time. 
2. Design custom CUDA functions that are 3x faster than the CUDA API.
3. Implement the nonlinear optimization problem which involves the math about Linear algebra and calculus. 
4. Work with backend teams to push products to market and build QA pipelines.
3D Animation:
1. Skinning the 3D models and automatically animate the 3D models to the arbitrary pose.

Title: 3D Human Body Generator and Cloth Generation
1. The project is my master thesis trying to solve virtual try-on problems. 
2. Propose a lite model generating 3D human body which is suitable for web page and mobile device, and the running time is less than 1 ms. 
3. Build the first pipeline for 3D cloth generation which can create proper training data for neural network, and the pipeline conforms to fashion industry.  

Video:
https://www.youtube.com/watch?v=dVLWum9r-3k&ab_channel=Che-KueiChang

Thesis Work

1. Detect the contour of the knee bone in X-ray and find the corresponding feature point from X-rays and 3D bone template.
2. Propose a graph algorithm to deform the template bone model to match the X-rays data. 
3. The software can potentially save patients 100s of dollars and weeks of waiting for CT scan. The patient can receive their diagnosis as soon as taken X-rays. 
4. We sold the software to Wind Intelligence Corp. as technology transfer.


Wind Intelligence Corp

Software Engineer, Tech Transfer

1. Trace and debug the external project code, then analyze the data structures and running time of each code segment.
2. Detect the face in the digital image and map it onto the 3D avatar model.
3. Design the algorithm and speed up 50% running time. 


Partager