Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/mateuszk098/car-brand-recognition
End to end machine learning project with full stack application for car brand recognition.
https://github.com/mateuszk098/car-brand-recognition
deep-learning docker fastapi full-stack machine-learning python pytorch
Last synced: about 2 months ago
JSON representation
End to end machine learning project with full stack application for car brand recognition.
- Host: GitHub
- URL: https://github.com/mateuszk098/car-brand-recognition
- Owner: mateuszk098
- License: mit
- Created: 2024-09-10T20:02:16.000Z (4 months ago)
- Default Branch: master
- Last Pushed: 2024-10-23T20:50:47.000Z (2 months ago)
- Last Synced: 2024-10-24T08:43:10.952Z (2 months ago)
- Topics: deep-learning, docker, fastapi, full-stack, machine-learning, python, pytorch
- Language: Python
- Homepage:
- Size: 111 KB
- Stars: 0
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Car Brand Recognition
 
## 1. Introduction
This project is a FastAPI-based web application that allows users to upload car images and predict the car brand using a custom Convolutional Neural Network (CNN) model. The model leverages **Residual** and **Squeeze-and-Excitation** connections to enhance prediction performance. It supports predictions for **25 car brands**, and users can specify how many top predictions they would like to see.
## 2. Key Features
- **Custom Deep Learning Model**: Built with lazy layers, including **Residual** and **Squeeze-and-Excitation** blocks, the model can accurately classify cars from a selection of 25 brands.
- **Multiple Car Brands Supported**: Audi, BMW, Toyota, and more...
- **Top K Predictions**: Users can request the top K most probable car brands, allowing for detailed feedback on possible matches.
## 3. Authentication and Authorization
- **User Registration & Login**: The application has full authentication and authorization mechanisms in place. Users must create an account and log in before submitting images for prediction.
- **Secure Access**: The app uses OAuth2 with JWT for token-based authentication, ensuring that only authenticated users can access the car brand prediction service.
## 4. How it Works
1. **Upload an Image**: Users can upload an image of a car.
2. **Specify Number of Predictions**: The user can choose how many top predictions they would like to retrieve.
3. **Receive Results**: The app returns a list of predicted car brands with their corresponding probabilities.
## 5. Technology Stack
- **FastAPI**: For building the web API and handling user requests.
- **Custom CNN**: A convolutional neural network enhanced with **Residual** and **Squeeze-and-Excitation** blocks to improve classification performance.
- **Authentication**: OAuth2-based token system for secure user management.
## 6. Model Performance Metrics
The following table summarizes the key performance metrics for the car brand classification model:
| | Accuracy | Precision | Recall | F1 Score |
|--------------|----------|-----------|--------|----------|
| **Value** | 79.18% | 80.26% | 79.18% | 79.25% |
## 7. Quick Start
1. Go to `car-brand-recognition/` directory.
2. Create `.env` file. It should look as follows:
```bash
# GENERAL
DEBUG = True
# MLFLOW POSTGRES
MLFLOW_POSTGRES_PORT = 5432
MLFLOW_POSTGRES_USER = "admin"
MLFLOW_POSTGRES_PASSWORD = "admin"
MLFLOW_POSTGRES_DB = "mlflow"
MLFLOW_POSTGRES_HOSTNAME = "mlflow-db"
# MLFLOW
MLFLOW_PORT = 5000
MLFLOW_TRACKING_URI = "http://localhost:${MLFLOW_PORT}"
MLFLOW_BACKEND_STORE_URI = "postgresql://${MLFLOW_POSTGRES_USER}:${MLFLOW_POSTGRES_PASSWORD}@${MLFLOW_POSTGRES_HOSTNAME}:5432/${MLFLOW_POSTGRES_DB}"
# PGADMIN
PGADMIN_PORT = 80
PGADMIN_DEFAULT_EMAIL = "[email protected]"
PGADMIN_DEFAULT_PASSWORD = "admin"
# APP POSTGRES
APP_POSTGRES_PORT = 5433
APP_POSTGRES_USER = "admin"
APP_POSTGRES_PASSWORD = "admin"
APP_POSTGRES_DB = "brands"
APP_POSTGRES_HOSTNAME = "app-db"
# FASTAPI
MODEL = "SEResNet3"
BACKEND = "openvino"
# Create secret key, for example by running: openssl rand -hex 32
JWT_SECRET_KEY = "d0d88f5cb30b4a4fb0982d3c3210f664f0ed49546bfc1c8ec09f1c9a334a7c47"
JWT_ALGORITHM = "HS256"
APP_DB_URL = "postgresql://${APP_POSTGRES_USER}:${APP_POSTGRES_PASSWORD}@${APP_POSTGRES_HOSTNAME}:5432/${APP_POSTGRES_DB}"
```
3. Go to `build/` directory and run `docker compose --env-file ../.env up -d`
4. Go to `http://localhost:8000/docs` and voilà.
## 8. Additional Steps
- If you want to experiment with the repository, create a virtual environment and run: `pip install -r requirements.txt`.
- If you don't want to run the application in a container because, for example, you want to experiment with it, comment out `APP_DB_URL` in the `.env` file. Then go to the `app/` directory and run: `fastapi dev main.py` This way, the application will be running directly on the host machine, and every change will be automatically considered.
- If you want to keep an eye on the database, go to `http://localhost:80` and login to pgAdmin 4 using `PGADMIN_DEFAULT_EMAIL` and `PGADMIN_DEFAULT_PASSWORD` that you set in `.env` file. Then, register a new server, you can name it "App". Connection host is `APP_POSTGRES_HOSTNAME` and the password is `APP_POSTGRES_PASSWORD`. If you want to experiment further with the application on the machine host, but subsequently running database on container, change: `{APP_POSTGRES_HOSTNAME}:5432` to `localhost:${APP_POSTGRES_PORT}` in `APP_DB_URL`.
- If you want to trace changes in SQLite database, you can use: `sqlite3 brands.db`, after running the application.