Chia Wei Wu

Hsinchu, Taiwan.

I am a second-year graduate student at National Tsing Hua University (NTHU) Vision Science Lab advised by Prof. Min Sun. I am interested in the application of computer vision from research to production. My current project focuses on design an Human Activity Analysis System in the factory with light-weight detector. Beside, my current research lays in unsupervised domain adaptation (UDA) and semi-supervised learning (SSL) for real-time object detection. Before that, I worked on knowledge distillation for object detection.


National Tsing Hua University Vision Science Lab

Team Leader to build an AI-assited efficiency analysis system by human activity in POC for Mirle

  • Design a Multi-Process System with 16 cameras requirements to automatically distribute video from different channels for inference and combine information from multiple videos for event analysis to assist the factory in solving efficiency problems.
  • Propose an Active Sampling Pipeline to obtain important data from daily videos for subsequent optimization by analyzing error cases in the detection output and motion flow at different phases. The current result is a 33% reduction in labor costs.
  • Propose an Auto Training Pipeline to continuously optimize the detector for the online system with datasets from active sample.
  • Design an Event Generator by analyzing human and machine activity to generate event reports. We propose a method for reconstructing occluded objects that reduces the false positive rate by 57% with the performance of 0.94 recall score.
  • Developed the segmentation and event labeling tools for more efficient annotation of privacy-preserving images.
  • Team member to optimize and apply an AI-assisted home care system on Edge AI

  • Proposed method for optimizing the YOLO detector to achieve a 0.91 recall score in a fisheye camera with fewer real fall images.
  • Designed an Auto Training Pipeline for reevaluates and redeploys appropriate model weight for each camera within 3 days.
  • Developed the part of presentation and notification of event information with AWS service and submit the iOS APP.
  • March 2022 — Present

    Multimedia Information System Lab

    Undergraduate Researcher to optimize and deploy a light-weight detector in mobile device for Bovia

  • Implemented the approach of knowledge distillation to optimize the 0.8M parameter YOLO detector achieves 0.52 mAP score.
  • Deployed the model by Tencent (TNN) framework in android cellphone (Xiaomi A2) with a size of 2.7MB and speed of 8.9 FPS.
  • Undergraduate Researcher to build a client-server application of smart home care system for Industrial Technology Research Institute (ITRI) and Taipei Veterans General Hospital (TVGH)

  • Designed the real-time recognition system (OD-RASH) to obtain blood pressure values and emergency notifications.
  • Optimized the ResNet classifier to classify the quality of captured image for users and YOLO detector to obtaion blood pressure values deployed in server which achieves 98% accuracy and 0.99 precision scores, separately.
  • Designed an iOS application to deploy the classifier model by PyTorch Mobile for real-time help users to capture a high-quality photo in iPhone XS with 20 FPS then communicate with server system to receive the emergency notifications.
  • Sep 2020 — June 2021


    Programming Languages & Tools
    Research Domains
    • Computer Vision
    • Deep Learning
    • Machine Learning
    • Cooperation Project

    • Sort out the key points
    • Sort the pain points
    • Distribute work
    • Unit test


    National Tsing Hua University

    Master of Science in Electrical Engineering

    Advisor: Prof. Min Sun

    Relevant Coursework:

    FinTech, Natural Language Processing Lab, Net Art, ML, Music Information Retrieval

    Sep 2021 — (Excepted) June 2023

    Yuan Ze University

    Advisor: Prof. Duan-Yu Chen

    Bachelor of Science in Electrical Engineering
    Relevant Coursework:

    Neural Network, DL, ML, Medical Image Processing

    Sep 2016 — June 2021


    Machine Learning Engineer

    Software Engineer

    Activity & Awards & Certifications

    • Word Recommendation System for Movie Reviews (course: Natural Language Processing Lab) , NTHU Feb 2022 - Jun 2022

    • Designed the system for film critics to get word recommendations in real-time by optimized the BERT-based model and apply NLTK and Hugging Face modules.
    • Auto Analysis Tech Stocks Implementation (course: Financial Technology), NTHU Sep 2021 - Jan 2022

    • Used web crawlers to receive the latest global news and optimize the regression model to predict stock trends.
    • Workshop on Synopsys ARC CPU with TensorFlow Lite, NTHU Apr 2021 - Apr 2021

    • Global Wheat Detection in Kaggle Competitions Jul 2020 - Aug 2020

    • Proposed the training framework and reached the score equivalent to the sliver medal on the private leaderboard.
    • Synopsys ARC AIoT Winter Camp, NCTU (Hackathon Competition) Feb 2020 - Feb 2020

    • Developed the functions of self-propelled vehicle combined with AI on the embARC platform of the ARC IoT Development Kit.
    • Traditional Image-Process Implementation (audition course: Fundamentals of Multimedia) Sep 2019 - Jan 2020
    • Implemented seam carving, watermark, and image retrieval by the traditional image-processing method.