Wenguang Wang’s Homepage
I’m Wenguang Wang, a Master in the Institute of Computer Science & Technology of Peking University (ICST), also a member of Character Shape Computing Lab (CSCL).
Education
- Master: 2017.09-2020.07 Peking University Computer Science
- Bachelor: 2013.09-2017.07 Sichuan University Electronic Engineering
Skills
- Skilled at c++, python, matlab, tensorflow.
- Familiar with machine learning and deep learning algorithms.
Project Experience
Chinese Stroke Extraction by Semantic Segmentation and Tabu Search
- Task: stroke extraction of Chinese character.
- Two improvements in original semantic segmentation models (Such as DeepLab v1, v3+):
- Stroke list module: guide the model by priori information like stroke count.
- Inference module: replace the background channel by binary images of the characters.
- Use tabu search algorithm for stroke ordering problem. We regard it as a Traveling Salesman Problem (TSP) and add stroke categories penalty term into the fitness function to improve the accuracy.
- Build a dataset containing 10 different fonts for stroke extraction (Pascal voc 2012 style).
- 300 times faster and 4.65% more accurate than the state-of-the-art.
Awards
- National Scholarship (twice, 2014–2015&2015–2016)
- Tanglixin Scholarship (60 people per year, 2016–2020)
- Undergraduate Electronic Design Contest, the Second Prize (international)
wangwenguang at pku dot edu dot cn