Traffic Sign Classifier

Description

A single-page web app serving a machine learning model which classifies traffic signs.

This Flask app is served by Gunicorn with NGINX as the reverse proxy.

The Machine Learning model is a Convoluted Neural Network (CNN) trained by myself using the TensorFlow library. See here for more details on how I trained this CNN.

Getting Started

1. Visit the web application here.

Video Demo

Traffic.Sign.Classifier.Demo.mp4

Specification

• Design and build a single-page web app to serve a Machine Learning model I trained.
• Serve the Python Flask app with Gunicorn and use NGINX as the reverse proxy. Customise NGINX config to force SSL.
• Serve the TensorFlow model using TensorFlow Serving for optimal performance and managing model versioning.
• Flask, NGINX and TensorFlow Serving Docker containers are orchestrated using Docker Compose.
• Users can upload their own images or select a random image for the model to classify.
• Uploaded images are not stored by the web app. Files are checked to determine if they are valid images.
• Dismissible messages are displayed to the user to notify machine learning inference success or error.
• Front-end is styled using the Bootstrap CSS framework, crafting a mobile-friendly layout.
• Using JavaScript, users can search and sort the information displayed in the table.

Technologies Used

• Docker with Docker Compose
• TensorFlow with TensorFlow Serving
• Python with Flask framework & OpenCV library
• Gunicorn
• NGINX
• HTML5
• CSS with Bootstrap framework
• JavaScript with DataTables library
• GitHub Actions workflows for CI to automate running tests via Docker Compose
• Dependabot to keep dependencies up to date and mitigate security vulnerabilities
• Render for deployment

Running locally

1. Ensure you have Docker installed.
2. To start the app, in the root folder where docker-compose.yml is contained, run docker-compose up in the terminal.
3. Visit localhost:1337 in your web browser. Bypass the invalid SSL certificate warning. E.g. in Chrome, click on the screen with the page open and type thisisunsafe to bypass the warning.
4. If you are unable to bypass the warning, checkout the non-ssl branch instead to run the Docker containers without SSL enforced. In this branch, visit localhost instead.
5. To shut down the app, run docker-compose down in the terminal or hit CTRL+C.

Testing

To run the integration tests, in the root folder, run the following in the terminal:

sh tests/test.sh