Adam Wang
2+ years of experience in Computer Vision, Deep Learning.
6+ years of experience in Python programming.
Professional in DeepFake Detection, AI OCR.
Taipei, Taiwan
[email protected]
Thesis
Face Forgery Detection with Frequency and Recurrent Features
TF-Net is a deepfake detection model, considering features from self-supervised template maps, frequency domain by the DCT (discrete cosine transform) module, sequential correlation from the GRU module.
This work has the SOTA performance on generalization of Celeb-DF dataset. (0.85 AUC score)
Work Experience
Software Engineer Aug. 2021 -- Nov. 2021
Sightour, Inc.
I am responsible for training multiple handwritten OCR models, adopting and improving current deep learning-based SOTA handwritten generating technology to generate handwritten data, designing specific post-rule for each type of insurance sheet and telecom bills to increase the accuracy of the word for word recognition rate by 30%.
Crawler Engineer Intern Dec. 2020 -- Jul. 2021
Big Data Co., Ltd.
As a member of the crawler team, I'm responsible for the development and maintenance of the crawler system of several thousands of websites, blogs, news, and social media (fb, ig), crawling by Python Scrapy framework, parsing documents by PyQuery, storing them into Redis or Elasticsearch database, testing by Pytest.
Research Intern Jul. 2020 -- Aug. 2020
Research Center for Information Technology Innovation, Academia Sinica
My team is responsible for Image Forgery Detection related research, studying SOTA papers and improving its algorithm, adding attention, non-local, ASPP, guiding mask and other mechanism, and combining other common image segmentation algorithms in order to achieve performance beyond that of the SOTA method.
Skills
- Languages: Python, R, C++, C, JavaScript.
- Frameworks and Tools: PyTorch, TensorFlow, Keras, Scrapy, Selenium, Git, Vim, Tableau, Docker, Shell.
- Web Technologies: D3.js, HTML, XML, SQL, Redis.
Education
National Taiwan Normal University Sep. 2019 -- June 2021
- M.S. in Computer Science and Information Engineering
Soochow University Sep. 2015 -- June 2019
- B.S. in School of Big Data Management
Projects
Scrapy Crawler and Flask Data Visualization
Static website crawler of PTT Gossiping forum using Scrapy framework, processing and analyzing data , visualizing popular articles and comments on website by Flask framework.
Irwin Mango Grade Classification Competition
Use Yolov4 for mango area detection, cropping and fed to some common image recognition backbones (such as AlexNet, HRNet, etc.) to learn defect features, and then use ensemble method, deep mutual learning for grade classification.
Web Design and Data Connection
Use D3.js to implement website data visualization, use php to access MySQL database, and draw the US map with some skills like transition, and hover with a variety of visual channels to connect a histogram with a map of the USA.
Final Project of Pattern Recognition Class
Fusion and modify two CVPR 2019 paper's algorithms ArcFace and RegularFace to reach 75% accuracy for face recognition on CASIA-WebFace dataset, which has 0.5M images for 7927 identities.
Graduation Project
SlopeOne is a general algorithm of recommender systems, which follows the user's usage record and compares the user's possible purchase behavior with other similar records in the database. This research combines the dynamic K-nearest neighbor theory to solve the common scalability problems of recommender systems.
Additional Information
Languages: Native in Chinese, fluent in English (TOEIC: 810/990), beginner in Korean.
