Bio. I am a computer science M.Eng. student at Cornell. I love to develop applications involving image processing, computer vision, and deep learning. My past projects include an anti-piracy web application using imperceptible image watermark, a deep reinforcement learning package in torch7 and an implementation of structural similarity (SSIM) index in pytorch. I am currently working on an affordable human pose capture tool using a single webcam.

Before I came to the US, I received MS and BS degree in electrical engineering from National Taiwan University. I published my MS research about camera raw data image processing on ICIP 2015 and got an internship at Himax Imageing Inc..

Timeline.
2017-2018: Cornell Computer Science M.Eng. student
Since 2017: Founder of PatreonMarker An anti-piracy web tool for creators on Patreon.
2016-2017: Compulsory Military service
Winter 2016: Acadamic Sinica Research Assistance Learning camera autofocus with reinforcement learning. Adviser: Yu-Chiang Frank Wang.
Summer 2015: Himax Imaging Inc. Internship Image processing algorithm for RGB-Infrared cameras to improve iris recognition
2013-2015: National Taiwan University: MS Compensation of Spectral Mismatch to Enhance WRGB Demosaicking. Adviser: Homer H. Chen
2009-2013: National Taiwan University: BS Electrical Engineering. Ranking: 43/204

Projects

PatreonMarker
PatreonMarker is an anti-piracy file hosting service for creators at Patreon. It sends each consumer a unique watermarked content so that when a consumer redistributes his own content, the creator can identify the consumer with the content.
dprl: deep reinforcement learning in torch7
dprl is a user-friendly deep reinforcement learning package for torch7. It contains an implementation of Google Deepmind's state-of-the-art algorithms: deep Q-learning, double deep Q-learning and asynchronous advantage actor critic.
Pytorch SSIM
pytorch-ssim is a pytorch implementation of the well-known Structural similarity (SSIM) index. With pytorch's powerful automatic differentiation, we can now develop algorithms that optimize SSIM instead of mean square error. The gif image shows the process of optimizing a noise image to the original image.

Publication

Compensation of Spectral Mismatch to Enhance WRGB Demosaicking,
Po-Hsun Su, Po-Chang Chen, and Homer H. Chen
ICIP 2015