A fun web application that can read one or more faces and pinpoint them with high accuracy on any number of subsequent test images. Uses a Next.js frontend and a Spring Boot backend.
Currently, the web interface only allows one target face and one test image to be uploaded but I'm working on that.
After following setup instructions in the Dependencies & Setup section, then with the Next.js and Spring Boot servers running:
- Visit localhost:3000 where you'll be greeted with a simple, very carefully thought out, well-made interface (thank you Lawrence).
- Upload an image of your face or some target face that you want the program to pinpoint in the following test image.
- Upload a test image, which the app will attempt to identify said face within.
The app will spit out a result image, which is the test image with a red bounding box drawn around the target's face. If the target face can't be found in the test image, the test image is just rendered with no modifications.
- Node.js 18.17+
- Java Development Kit 8.x
- Maven 3.x (optional)
It's worth mentioning that I use pnpm to manage frontend packages and to run the frontend, although I don't know whether it's actually required to run this project.
Install the necessary node_modules by running npm install inside the frontend Next.js project, or whatever the pnpm equivalent is. I run pnpm dev to run the frontend server (I'm not a frontend guy).
For the backend Spring Boot project, running mvn package (or ./mvnw package if you don't have Maven) should be all that's required. After that, Application.java can be executed to start up the backend server.
This app utilizes pre-trained YuNet face detection and SFace face recognition deep neural network models, which are present in this repository.
Thank you Professor Shiqi Yu and Yuantao Feng for training and providing the YuNet model!
Thank you Professor Weihong Deng, PhD Candidate Zhong Yaoyao, and Master Candidate Chengrui Wang for training and providing the SFace model!
Initially developed as a prototype, this project was designed for possible integration into the company where I completed my internship in Summer 2024. Before I started looking at OpenCV, I asked my office manager what programming language I should use for the application and the answer was "preferably Java".
This made sense in the context of the workplace because most of the company's projects were written in Java, allowing seamless integration. However, this did not make sense in the context of OpenCV as I discovered quickly that OpenCV is native to C++. While there is an official OpenCV Java API, I chose to use a third-party API mostly because the official one isn't available as a Maven dependency.
I say Java makes little sense because I could have easily done this project in C++ or Python, as Java was only ever a recommendation. But ever since I started writing the code, I kept on gaslighting myself into thinking that I was in too deep already.
As a result, I dealt with scarce documentation (having to cross-reference C++ docs), unhelpful error messages (if C++ errors weren't cryptic enough, they look worse coming from the JVM), and a LOT of manual memory management (why do I have to free C pointers in Java?).
But hey, the program works so the end justifies the means.