Kyle Shih-Huang Lo

Kyle Lo's Portfolio

Machine Learning

An Improved Video2mesh Tool based-on PIFu (2022)

Automatically segments foreground from video frames using MobileNetV3 and feeds foreground masks and frames into PIFu to generate a series of meshes
Used Pytorch3D, MediaPipe, OpenCV, Python, Shell Script
Please find the source code [here]

Music Genre CNN Classifier (2018)

Built a convolutional neural network to classify (non-)progressive rock music and with kernel visualizations
Used Google Colab, TensorFlow, scikit-learn
Please find the source code [here]

Image & Video Processing

Visual-Quality Guided Global Backlight Dimming for Video Display on Mobile Devices (2018)

Analyzed images and considered characteristics of human vision to adjust frames & LCD backlight to achieve power saving on mobile devices without affecting visual quality.
Used Android NDK, Open GL ES, JAVA, and MATLAB.
The first part of Kyle's master thesis, and has been published to IEEE TCSVT. [PDF]

Visual-Attention-Based Pixel Dimming Technique for OLED Displays of Mobile Devices (2018)

Considered saliency map generated from content motion to dim pixel brightness locally on mobile devices' OLED display while maintaining good visual quality.
Used Android NDK, Open GL ES, JAVA, and MATLAB.
VAB-PD is the second part of Kyle's master thesis and has been published to IEEE TIE. [PDF]

IC Crack Detection (2016)

Built an automatic IC defeat detection system to reduce workloads for inspectors.
Used image enhancement and denoising techniques to strengthen the features of cracks; then, utilized pattern recognition and SVM method to extract cracks.
Industry-academy cooperation project with ASE Group. [PDF]

3D Modeling and Interaction

Polyhedral Splines Modeling Addon for Blender (2021)

Developed a Blender addon that helps users generate B-spline surfaces from arbitrary meshes and enable real-time editing
Used Python & Blender API
Please visit the intro website and the repo

All the Ways to Drunk (2015)

Used wine bottle breaking animation to metaphor the tragedy caused by drunk driving.
Used MAYA & RealFlow to model the 3D scene, and utilized VRAY to finish the texture rendering.
Three people worked on this animation from 2015.03 to 2015.09, and Kyle was in charge of the liquid simulation.

3D Virtual Strength Training System (2015)

Built a scene with virtual-physical therapists to guide patients rehabilitate at home via Microsoft Kinect.
Used MAYA to model 3D scenes; utilized Kinect SDK and Unity to build the body pose tracking and guiding system.
Industry-academy cooperation project with HIWIN Crop.

Course Project

Monkey Alarm System

Built an alarm system for students to report monkey's location on campus to prevent everyone's food from robbing.
Used Android and Google App Engine.
A final project in Object-oriented Programming course.

Special Certification

Four-Dan in Go Game

GO is the most complex board game in the world. [wiki]
Since Go has the deepest decision tree of all chess games, players are required to have strong logical thinking and reasoning ability.
Kyle has been certificated for Four-Dan in Go ranking via the participation in national open for Go games.

Works of Arts

Lotus & Frogs (2003)

White Eyes (2006)

Snow Path (2004)

Rock Fish Falls (2005)

Heaven Fall (2001)

Dream (2005)

Maple Fire (2002)

Thanks (2004)

Google Sites

Report abuse