}

Introduction

Welcome to my personal website! I am currently an undergraduate fourth year student at Georgia Institute of Technoloy studying Computer Engineering and Industrial Design. I enjoy coding and am currently learning about computer vision. I recently spent the summer participating in a National Science Foundation program doing a web implementation of a real-time object detection model called YOLO-LITE. For more information, check out the link below:

YOLO-LITE website
Rachel Huang

Contact Me

Institution: Georgia Institute of Technology

Major: Computer Engineering

Minor: Industrial Design

Favorite Langauges: C, C++, Java, Python

rachuang22@gmail.com

LinkedIn

GitHub

National Science Foundation: Data Mining and Statistics

ABOUT

Where: University of North Carolina Wilmington

What: National Science Foundation, Research Experience for Undergraduates

When: Summer 2018

Project: YOLO-LITE: A web implementation of real-time object detecion.

Click here for the Live Demo.

Click here for our GitHub.

Skills Earned:

R
Python
Machine Learning
LaTex
HTML, CSS, JS
My dog, Rain, after being classified by our YOLO-LITE model.

I would like to give a special thanks to Jonathan Pedoeem for collaborating on this project and Dr. Cuixian Chen + Dr. Yishi Wang for guiding and mentoring me through this summer program. For more information, please visit Dr. Chen's website.

Timeline

Week 1: During the first couple of weeks I mainly worked with a large dataset compromised of mugshots(MORPH-II). I learned data mining techniques and how to clean the dataset in R. I applied some gender, age, and race classification such as LDA, QDA, KNN, and so on. The BIF features from the dataset were too big to fully run on R. Week 1 Presentation

Week 2: I continued to learn about classification methods such as regression trees, bootstrapping, bagging, random forest, and boosting. During this week I decided to learn switch to Python in preparation for my research in deep learning. Week 2 Presentation

Week 3: While week one and two were dedicated to data mining, week three moved to dimension reduction techniques such as LDA, KLDA, PCA, KPCA, and MDS. Most of the projects were re-running classification and regression techniques from week one and two after dimension reduction techniques were applied. Week 3 Presentation

Week 4: After an introduction to statistics from week one-three, I finally began to formulate my project for the summer. I mostly read research papers on neural networks in order to gain a better understanding of some ideas for a research topic. In the end, I decided to focus my research on a real-time object detection web implementation called YOLO-LITE. Week 4 Presentation

Week 5: I spent most of this week trying to implement an already written object detection model called YOLO (You Only Look Once). I was able to successfully implement a real-time object detection (pictured on the left) running at about 2 FPS. My goal for the summer is to implement some efficient techniques in order to reach around 10 FPS. Week 5 Presentation

Week 6: The REU program received a GPU for our training purposes. After dedicating a couple days to get the GPU running, we ran into some problems with training our models. The Mean Average Precision of our retrained tiny-YOLO-VOC model was around 0.5% while the already trained model was around 50%. With our MAP being so low, we decided to switch from using Python and Tensorflow to C. Week 6 Presentation

Week 7: Switching to the original C implementation of YOLO, training began working successfully. The program received yet another GPU which helped speed up the training process a bit. We were able to achieve 21 FPS at 30% Mean Average Precision with training on PASCAL VOC dataset. (No presentation this week)

Week 8: I spent most of this week moving our model from C to java script in order to implement it onto a website. Although there were some problems with dependencies, our best models trained on MS COCO and PASCAL VOC were pushed live onto the website. Week 8 Presentation

Week 9: Week 9 was a wrap up week in preparation for the final poster session, final presentation, and final paper in week 10. I wrote a python script in order to run all the weights files from training to find the best mAP for each model. We also updated some pictures on our website and pushed all our cfg and weights files to our GitHub. (No presentation this week.)

Week 10: The final week of the program was dedicated to finishing our final poster, paper, and presentation. The links are provided to the left.

Industrial Design: Visual Design Thinking

EXPLODED NACHOS

Assignment: Draw your favorite food in an exploded view.

Explanation: This was an assignment that was assigned for the Introduction to Visual Design Thinking class I took Spring 2020. The assignment was to draw an exploded view of a food of my choice. Exploded view is a drawing that shows the relationship or order of various parts of an object.

I chose nachos because nachos have many different layers when being represented in an exploded view. All of the layers have vertical relationship with each other.

PIZZA DELIVERY

Assignment: Design a delivery robot.

Explanation: This was the final project for the Introduction to Visual Design Thinking class I took Spring 2020. The goal of the project was to combine all the different techniques we learned in class into a final design.

I designed a pizza robot inspired by the Domino's pizza delivery robot. My robot is made to be friendly and approachable. When the user first interacts with the bot, the glass panel holding the pizza and the drinks will be locked until the user pays with the credit card system on the side. Once payment is approved, the robot will lift the glass and present the pizza and/or drinks. It moves using a ball on the bottom to roll and move with smooth motion.

Personal Projects

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.

Stay tuned for more.