Hello!

I received my M.S. degree in the Department of Computer Science at National Taiwan University in 2017, supervised by Prof. Yung-Yu Chuang and Dr. Yen-Yu Lin from Academia Sinica, where I worked as a part-time research assistant. Before these, I received my B.S. degree in the Department of Department of Computer Science at National Taiwan University.

My research interests are computer vision and deep learning.

Please find my latest resume here.

News

Sep. 2017: Received Excellent Master Thesis Award from Image​ ​Processing​ ​and​ ​Pattern​ ​Recognition​ ​Society

Aug. 2017: Received M.S. in Computer Science, National Taiwan University

Feb. 2017: One paper is accepted in CVPR 2017!

Publication

Deep Co-occurrence Feature Learning for Visual Object Recognition

Ya-Fang Shih*, Yang-Ming Yeh* (* indicates equal contribution), Yen-Yu Lin,
Ming-Fang Weng, Yi-Chang Lu, Yung-Yu Chuang
IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017

Project

Stereo Panorama

Stereo image pairs (for left and right eyes) are that people can perceive 3D stereo by looking at the corresponding one for each eye. This system produces stereo panorama image pairs from a hand-held GoPro video. The work focuses on implementing the flow-based image blending method proposed by Megastereo by Richardt et al and uses the omnistereo method to extract the image strips for each eye.

Outfit Color Harmony Evaluation System

Does these clothes go well for me to wear? This outfit evaluation system scores how harmonic the colors of people’s outfit looks. The work focuses on implementing the color harmonization algorithm proposed by Cohen-Or et al.

Distorted Movie Scene Image Classification

Photos taken by users in real life are usually different from the training data we have, and are often with heavy lightning and contrast distortion. This work focuses on improving the classification accuracy of the photos of movie scenes shot by users, which have heavy distortion. Developed with MatConvNet.

Image Feature Matching Android Application

The speed of matching feature between images on Android platfroms depends on the programming languages used, and it can sometimes take a long time. This work focuses on implementing feature matching using native language C++ and integrating it into the Java environment on Andriod systems. The resulting application takes photos, detects and matches robust feature points in pictures instantly on mobile devices. Developed with Android NDK and OpenCV4Android.

DJ Board

What does your motion sound like? DJBoard is an interactive skateboard that people can mix their own track of music on by performing different board-motions and triggering different sensors. Developed with an infrared sensor, pressure and ultrasonic sensors and a gyroscope on Arduino platform.